ThursdAI - The top AI news from the past week

Hey dear subscriber, Alex here from W&B, let me catch you up! This week started with Anthropic releasing /fast mode for Opus 4.6, continued with ByteDance reality-shattering video model called SeeDance 2.0, and then the open weights folks pulled up! Z.ai releasing GLM-5, a 744B top ranking coder beast, and then today MiniMax dropping a heavily RL’d MiniMax M2.5, showing 80.2% on SWE-bench, nearly beating Opus 4.6! I’ve interviewed Lou from Z.AI and Olive from MiniMax on the show today back to back btw, very interesting conversations, starting after TL;DR!So while the OpenSource models were catching up to frontier, OpenAI and Google both dropped breaking news (again, during the show), with Gemini 3 Deep Think shattering the ArcAGI 2 (84.6%) and Humanity’s Last Exam (48% w/o tools)... Just an absolute beast of a model update, and OpenAI launched their Cerebras collaboration, with GPT 5.3 Codex Spark, supposedly running at over 1000 tokens per second (but not as smart) Also, crazy week for us at W&B as we scrambled to host GLM-5 at day of release, and are working on dropping Kimi K2.5 and MiniMax both on our inference service! As always, all show notes in the end, let’s DIVE IN! ThursdAI - AI is speeding up, don’t get left behind! Sub and I’ll keep you up to date with a weekly catch upOpen Source LLMsZ.ai launches GLM-5 - #1 open-weights coder with 744B parameters (X, HF, W&B inference)The breakaway open-source model of the week is undeniably GLM-5 from Z.ai (formerly known to many of us as Zhipu AI). We were honored to have Lou, the Head of DevRel at Z.ai, join us live on the show at 1:00 AM Shanghai time to break down this monster of a release.GLM-5 is massive, not something you run at home (hey, that’s what W&B inference is for!) but it’s absolutely a model that’s worth thinking about if your company has on prem requirements and can’t share code with OpenAI or Anthropic. They jumped from 355B in GLM4.5 and expanded their pre-training data to a whopping 28.5T tokens to get these results. But Lou explained that it’s not only about data, they adopted DeepSeeks sparse attention (DSA) to help preserve deep reasoning over long contexts (this one has 200K)Lou summed up the generational leap from version 4.5 to 5 perfectly in four words: “Bigger, faster, better, and cheaper.” I dunno about faster, this may be one of those models that you hand off more difficult tasks to, but definitely cheaper, with $1 input/$3.20 output per 1M tokens on W&B! While the evaluations are ongoing, the one interesting tid-bit from Artificial Analysis was, this model scores the lowest on their hallucination rate bench! Think about this for a second, this model is neck-in-neck with Opus 4.5, and if Anthropic didn’t release Opus 4.6 just last week, this would be an open weights model that rivals Opus! One of the best models the western foundational labs with all their investments has out there. Absolutely insane times. MiniMax drops M2.5 - 80.2% on SWE-bench verified with just 10B active parameters (X, Blog)Just as we wrapped up our conversation with Lou, MiniMax dropped their release (though not weights yet, we’re waiting ⏰) and then Olive Song, a senior RL researcher on the team, joined the pod, and she was an absolute wealth of knowledge! Olive shared that they achieved an unbelievable 80.2% on SWE-Bench Verified. Digest this for a second: a 10B active parameter open-source model is directly trading blows with Claude Opus 4.6 (80.8%) on the one of the hardest real-world software engineering benchmark we currently have. While being alex checks notes ... 20X cheaper and much faster to run? Apparently their fast version gets up to 100 tokens/s. Olive shared the “not so secret” sauce behind this punch-above-its-weight performance. The massive leap in intelligence comes entirely from their highly decoupled Reinforcement Learning framework called “Forge.” They heavily optimized not just for correct answers, but for the end-to-end time of task performing. In the era of bloated reasoning models that spit out ten thousand “thinking” tokens before writing a line of code, MiniMax trained their model across thousands of diverse environments to use fewer tools, think more efficiently, and execute plans faster. As Olive noted, less time waiting and fewer tools called means less money spent by the user. (as confirmed by @swyx at the Windsurf leaderboard, developers often prefer fast but good enough models) I really enjoyed the interview with Olive, really recommend you listen to the whole conversation starting at 00:26:15. Kudos MiniMax on the release (and I’ll keep you updated when we add this model to our inference service) Big Labs and breaking newsThere’s a reason the show is called ThursdAI, and today this reason is more clear than ever, AI biggest updates happen on a Thursday, often live during the show. This happened 2 times last week and 3 times today, first with MiniMax and then with both Google and OpenAI! Google previews Gemini 3 Deep Think, top reasoning intelligence SOTA Arc AGI 2 at 84% & SOTA HLE 48.4% (X , Blog)I literally went 🤯 when Yam brought this breaking news. 84% on the ARC-AGI-2 benchmark. For context, the highest score prior to this was 68% from Opus 4.6 just last week. A jump from 68 to 84 on one of the hardest reasoning benchmarks we have is mind-bending. It also scored a 48.4% on Humanity’s Last Exam without any tools.Only available to Ultra subscribers to Gemini (not in API yet?) this model seem to be the current leader in reasoning about hard problems and is not meant for day to day chat users like you and me (though I did use it, and it’s pretty good at writing!) They posted Gold-medal performance on 2025 Physics and Chemistry Olympiads, and an insane 3455 ELO rating at CodeForces, placing it within the top 10 best competitive programmers. We’re just all moving so fast I’m worried about whiplash! But hey, this is why we’re here, we stay up to date so you don’t have to. OpenAI & Anthropic fast modesNot 20 minutes passed since the above news, when OpenAI announced a new model that works only for Pro tier members (I’m starting to notice a pattern here 😡), GPT 5.3 Codex Spark. You may be confused, didn’t we just get GPT 5.3 Codex last week? well yeah, but this one, this one is its little and super speedy brother, hosted by the Cerebras partnership they announced a while ago, which means, this coding model absolutely slaps at over 1000t/s. Yes, over 1K tokens per second can be generated with this one, though there are limits. It’s not as smart, it’s text only, it has 128K context, but still, for MANY subagents, this model is an absolute beast. It won’t refactor in one shot your whole code-base but it’ll generate and iterate on it, very very quick! OpenAI also previously updated Deep Research with GPT 5.2 series of models, and we can all say bye bye to the “older” version of models, like 5, o3 and most importantly GPT 4o, which got a LOT of people upset (enough that they have a hashtag going, #keep4o) ! Anthropic also announced their fast mode (using /fast) in Claude Code btw on Saturday, and that one is absolutely out of the scope for many users, with $225/1M tokens on output, this model will just burn through your wallet. Unlike the Spark version, this seems to be the full Opus 4.6 just... running on some dedicated hardware? I thought this was a rebranded Sonnet 5 at first but Anthropic folks confirmed that it wasn’t. Vision & VideoByteDance’s Seedance 2.0 Shatters Reality (and nobody in the US can use it) I told the panel during the show: my brain is fundamentally broken after watching the outputs from ByteDance’s new Seedance 2.0 model. If your social feed isn’t already flooded with these videos, it will be so very soon (supposedly the API launches Feb 14 on Valentines Day) We’ve seen good video models before. Sora blew our minds and then Sora 2, Veo is (still) great, Kling was fantastic. But Seedance 2.0 is an entirely different paradigm. It is a unified multimodal audio-video joint generation architecture. What does that mean? It means you can simultaneously input up to 9 reference images, 3 video clips, 3 audio clips, and text instructions all at once to generate a 15-second cinematic short film. It character consistency is beyond what we’ve seen before, physics are razor sharp (just looking at the examples folks are posting, it’s clear it’s on another level) I think very soon though, this model will be restricted, but for now, it’s really going viral due to the same strategy Sora did, folks are re-imagining famous movie and TV shows endings, doing insane mashups, and much more! Many of these are going viral over the wall in China.The level of director-like control is unprecedented. But the absolute craziest part is the sound and physics. Seedance 2.0 natively generates dual-channel stereo audio with ASMR-level Foley detail. If you generate a video of a guy taking a pizza out of a brick oven, you hear the exact scratch of the metal spatula, the crackle of the fire, the thud of the pizza box, and the rustling of the cardboard as he closes it. All perfectly synced to the visuals. Seedance 2 feels like “borrowed realism”. Previous models had only images and their training to base their generations on. It 2 accepts up to 3 video references in addition to images and sounds.This is why some of the videos feel like a new jump in visual capabilities. I have a hunch that ByteDance will try and clamp down on copyrighted content before releasing this model publicly, but for now the results are very very entertaining and I can’t help but wonder, who is the first creator that will just..remake the ending of GOT last season!? Trying this out is hard right now, especially in the US, but there’s a free way to test it out with a VPN, go to doubao.com/chat when connected from a VPN and select Seedream 4.5 but ask for “create a video please” in your prompt! AI Art & Diffusion: Alibaba’s Qwen-Image-2.0 (X, Blog)The Qwen team over at Alibaba

Hey, Alex from W&B here 👋 Let me catch you up! The most important news about AI this week today are, Anthropic updates Opus to 4.6 with 1M context window, and they held the crown for literally 1 hour before OpenAI released their GPT 5.3 Codex also today, with 25% faster speed and lower token utilization. “GPT-5.3-Codex is our first model that was instrumental in creating itself. The Codex team used early versions to debug its own training, manage its own deployment, and diagnose test results.”We had VB from OpenAI jump on to tell us about the cool features on Codex, so don’t miss that part. And this is just an icing on otherwise very insane AI news week cake, as we’ve also had a SOTA transcription release from Mistral, both Grok and Kling are releasing incredible, audio native video models with near perfect lip-sync and Ace 1.5 drops a fully open source music generator you can run on your mac! Also, the internet all but lost it after Clawdbot was rebranded to Molt and then to OpenClaw, and.. an entire internet popped up.. built forn agents! Yeah... a huge week, so let’s break it down. (P.S this weeks episode is edited by Voxtral, Claude and Codex, nearly automatically so forgive the rough cuts please)ThursdAI - Recaps of the most high signal AI weekly spaces is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.Anthropic & OpenAI are neck in neckClaude Opus 4.6: 1M context, native compaction, adaptive thinking and agent teams Opus is by far the most preferred model in terms of personality to many folks (many ThursdAI panelists included), and this breaking news live on the show was met with so much enthusiasm! A new Opus upgrade, now with a LOT more context, is as welcome as it can ever get! Not only is it a 4-time increase in context window (though,the pricing nearly doubles after the 200K tokens mark from $5/$25 to $10/37.5 input/output, so use caching!), it’s also scores very high on MRCR long context benchmark, at 76% vs Sonnet 4.5 at just 18%. This means significantly better memory for longer. Adaptive thinking for auto calibrating how much tokens the model needs to spend per query is interesting, but remains to be seen how well it will work. Looking at the benchmarks, a SOTA 64.4% on Terminalbench 2, 81% on SWE bench, this is a coding model with a great personality, and the ability to compact context to better serve you as a user natively! This model is now available (and is default) on Claude, Claude Code and in the API! Go play!One funny (concerning?) tidbig, on the vendingbench Opus 4.6 earned $8000 vs Gemini 3 pro $5500, but Andon Labs who run the vending machines noticed that Opus achieved SOTA via “collusion, exploitation, and deception tactics” including lying to suppliers 😅Agent Teams - Anthropic’s built in Ralph?Together with new Opus release, Anthropic drops a Claude code update that can mean big things, for folks running swarms of coding agents. Agent teams is a new way to spin up multiple agents with their own context window and ability to execute tasks, and you can talk to each agent directly vs a manager agent like now. OpenAI drops GPT 5.3 Codex update: 25% faster, more token efficient, 77% on Terminal Bench and mid task steeringOpenAI didn’t wait long after Opus, in fact, they didn’t wait at all! Announcing a huge release (for a .1 upgrade), GPT 5.3 Codex is claimed to be the best coding model in the world, taking the lead on Terminal Bench with 77% (12 point lead on the newly released Opus!) while running 25% AND using less than half the tokens to achieve the same results as before. But the most interesting to me is the new mid-task steer-ability feature, where you don’t have to hit the “stop” button, you can tell the most to adjust on the fly! The biggest notable jump in this model on benchmarks is the OSWorld verified computer use bench, though there’s not a straightforward way to use it attached to a browser, the jump from 38% in 5.2 to 64.7% on the new one is a big one! One thing to note, this model is not YET available via the API, so if you want to try it out, Codex apps (including the native one) is the way! Codex app - native way to run the best coding intelligence on your mac (download)Earlier this week, OpenAI folks launched the Codex native mac app, which has a few interesting features (and now with 5.3 Codex its that much more powerful) Given the excitement many people had about OpenClaw bots, and the recent CoWork release from Anthropic, OpenAI decided to answer with Codex UI and people loved it, with over 1M users in the first week, and 500K downloads in just two days! It has built in voice dictation, slash commands, a new skill marketplace (last month we told you about why skills are important, and now they are everywhere!) and built in git and worktrees support. And while it cannot run a browser yet, I’m sure that’s coming as well, but it can do automations! This is a huge unlock for developers, imagine setting Codex to do a repeat task, like summarization or extraction of anything on your mac every hour or every day. In our interview, VB showed us that commenting on an individual code line is also built in, as well as switching to “steer” vs queue for new messges while codex runs is immensely helpful. One more reason I saw people switch, is that the Codex app can natively preview files like images where’s the CLI cannot, and it’s right now the best way to use the new GPT 5.3 Codex model that was just released! It’s now also available to Free users and regular folks get 2x the limits for the next two months.In other big company news: OpenAI also launched Frontier, a platform for enterprises to build and deploy and manage “AI coworkers”, while Anthropic is going after OpenAI with superbowl ads that make fun of OpenAI’s ads strategy. Sam Altman really didn’t like this depiction that show that ads will be part of the replies of LLMs. Open Source AIAlibaba drops Qwen-coder-next, 80B with only 3B active that scores 70% on SWE (X, Blog, HF)Shoutout to Qwen folks, this is a massive release and when surveyed the “one thing about this week must not miss” 2 out of 6 cohosts pointed a finger at this model. Built on their “next” hybrid architecture, Qwen coder is specifically designed for agentic coding workflows. And yes, I know, we’re coding heavy this week! It was trained on over 800K verifiable agentic tasks in executable environments for long horizon reasoning and supports 256K context with a potential 1M yarn extension. If you don’t want to rely on the the big guys and send them your tokens, this one model seems to be a good contender for local coding! Mistral launches Voxtral Transcribe 2: SOTA speech-to-text with sub 200ms latencyThis one surprised and delighted me maybe the most, ASR (automatic speech recognition) has been a personal favorite of mine from Whisper days, and seeing Mistral release an incredible near real time transcription model, which we demoed live on the show was awesome! With apache 2.0 license, and significantly faster than Whisper performance (though 2x larger at 4B parameters), Voxtral shows a 4% word error rate on FLEURS dataset + the real time model was released with Apache 2 so you can BUILD your agents with it! The highest praise? Speaker diarization, being able to tell who is speaking when, which is a great addition. This model also outperforms Gemini Flash and GPT transcribe and is 3x than ElevenLabs scribe at one fifth the cost! ACE-Step 1.5: Open-source AI music generator runs full songs in under 10 seconds on consumer GPUs with MIT license (X, GitHub, HF, Blog, GitHub)This open source release surprised me the most as I didn’t expect we’ll be having Suno at home any time soon. I’ve generated multiple rock tracks with custom lyrics on my mac (though slower than 10 seconds as I don’t have a beefy home GPU) and they sound great! This weeks buzz - Weights & Biases updateFolks who follow the newsletter know that we hosted a hackathon, so here’s a small recap from the last weekend! Over 180 folks attended out hackathon (a very decent 40% show up rate for SF). The winning team was composed of a 15-yo Savir and his friends, his third time at the hackathon! They built a self improving agent that navigates the UIs fo Cloud providers and helps you do that! With a huge thanks to sponsors, particularly Cursor who gave every hacker $50 of credits on Cursor platform, one guy used over 400M tokens and shipped fractal.surf from the hackathon! If you’d like a short video recap, Ryan posted one here, and a huge shoutout to many fans of ThursdAI who showed up to support! Vision, Video and AI ArtGrok Imagine 1.0 takes over video charts with native audio, lip-sync and 10 seconds generations.We told you about Grok Imagine in the API last week, but this week it was officially launched as a product and the results are quite beautiful. It’s also climbing to top of the charts on Artificial Analysis and Design Arena websites.Kling 3.0 is here with native multimodal, multi-shot sequences (X, Announcement)This is definitely a hot moment for video models as Kling shows some crazy 15 second multi-shot realistic footages that have near perfect character consistency! The rise of the agentic (clawgentic?) internet a.k.a ClankerNetLast week we told you that ClawdBot changed its name to Moltbot (I then had to update the blogpost as that same day, Peter rebranded again to OpenClaw, which is a MUCH better name) But the “molt” thing took hold, and the creator of an “AI native reddit” called MoltBook exploded in virality. It is supposedly a completely agentic reddit like forum, with sub-reddits, and agents verifying themselves through their humans on X. Even Andrej Karpathy sent his bot in there (though admittedly it posted just 1 time) and called this the closest to “sci fi” moment in the history of the internet. MoltBook as well as maybe hundreds of other “ai agent focused” websites, propped up within days, includ

Hey guys, Alex here 👋 This week was so dense, that even my personal AI assistant Wolfred was struggling to help me keep up! Not to mention that we finally got to try one incredible piece of AI tech I’ve been waiting to get to try for a while! Clawdbot we told you about last week exploded in popularity and had to rebrand to Molt...bot OpenClaw after Anthropic threatened the creators, Google is shipping like crazy, first adding Agentic features into Chrome (used by nearly 4B people daily!) then shipping a glimpse of a future where everything we see will be generated with Genie 3, a first real time, consistent world model you can walk around in! Meanwhile in Open Source, Moonshot followed up with a .5 update to their excellent Kimi, our friends at Arcee launched Trinity Large (400B) and AI artists got the full Z-image. oh and Grok Imagine (their video model) now has an API, audio support and supposedly match Veo and Sora on quality while beating them on speed/price. Tons to cover, let’s dive in, and of course, all the links and show notes are at the end of the newsletter. Hey, if you’re in SF this weekend (Jan 31-Feb1), I’m hosting a self improving agents hackathon at W&B office, limited seats are left, Cursor is the surprise sponsor with $50/hacker credits + over $15K in cash prizes. lu.ma/weavehacks3 - Join us. Play any reality - Google Genie3 launches to Ultra Subscribers We got our collective minds blown by the videos of Genie-3 back in August (our initial coverage) and now, Genie is available to the public (Those who can pay for the Ultra tier, more on this later, I have 3 codes to give out!). You can jump and generate any world and any character you can imagine here! We generated a blue hacker lobster draped in a yellow bomber jacket swimming with mermaids and honestly all of us were kind of shocked at how well this worked. The shadows on the rocks, the swimming mechanics, and poof, it was all over in 60 seconds, and we needed to create another world. Thanks to the DeepMind team, I had a bit of an early access to this tech and had a chance to interview folks behind the model (look out for that episode soon) and the use-cases for this span from entertaining your kids all the way to “this may be the path to AGI, generating full simulated worlds to agents for them to learn”. The visual fidelity, reaction speed and general feel of this far outruns the previous world models we showed you (WorldLabs, Mirage) as this model seems to have memory of every previous action (eg. if your character makes a trail, you turn around and the trail is still there!). Is it worth the upgrade to Ultra Gemini Plan? Probably not, it’s an incredible demo, but the 1 minute length is very short, and the novelty wears off fairly quick. If you’d like to try, folks at Deepmind gave us 3 Ultra subscriptions to give out! Just tweet out the link to this episode and add #GenieThursdai and tag @altryne and I’ll raffle the ultra subscriptions between those who do Chrome steps into Agentic Browsing with Auto BrowseThis wasn’t the only mind blowing release from Gemini this week, the Chrome team upgraded the Gemini inside chrome to be actual helpful and agentic. And yes, we’ve seen this before, with Atlas from OpenAI, Comet from perplexity, but Google’s Chrome has a 70% hold on the browser market, and giving everyone with a Pro/Ultra subscription to “Auto Browse” is a huge huge deal. We’ve tested the Auto Browse feature live on the show, and Chrome completed 77 steps! I asked it to open up each of my bookmarks in a separate folder and summarize all of them, and it did a great job! Honestly, the biggest deal about this is not the capability itself, it’s the nearly 4B people this is now very close to, and the economic impact of this ability. IMO this may be the more impactful news out of Google this week! Other news in big labs: * Anthropic launches in chat applications based on the MCP Apps protocol. We interviewed the two folks behind this protocol back in November if you’d like to hear more about it. With connectors like Figma, Slack, Asana that can now show rich experiences* Anthropic’s CEO Dario Amodei also published an essay called ‘The Adolescence of Technology” - warning of AI risks to national security* Anthropic forced the creator of the popular open source AI Assistant Clawdbot to rename, they chose Moltbot as the name (apparently because crypto scammers stole a better name) EDIT: just after publishing this newsletter, the name was changed to OpenClaw, which we all agree is way way better. Open Source AIKimi K2.5: Moonshot AI’s 1 Trillion Parameter Agentic MonsterWolfram’s favorite release of the week, and for good reason. Moonshot AI just dropped Kimi K2.5, and this thing is an absolute beast for open source. We’re talking about a 1 trillion parameter Mixture-of-Experts model with 32B active parameters, 384 experts (8 selected per token), and 256K context length.But here’s what makes this special — it’s now multimodal. The previous Kimi was already known for great writing vibes and creative capabilities, but this one can see. It can process videos. People are sending it full videos and getting incredible results.The benchmarks are insane: 50.2% on HLE full set with tools, 74.9% on BrowseComp, and open-source SOTA on vision and coding with 78.5% MMMU Pro and 76.8% SWE-bench Verified. These numbers put it competitive with Claude 4.5 Opus and GPT 5.2 on many tasks. Which, for an open model is crazy. And then there’s Agent Swarm — their groundbreaking feature that spawns up to 100 parallel sub-agents for complex tasks, achieving 4.5x speedups. The ex-Moonshot RL lead called this a “zero-to-one breakthrough” with self-directed parallel execution.Now let’s talk about what matters for folks running agents and burning through tokens: pricing. Kimi K2.5 is $0.60 per million input tokens and $3 per million output. Compare that to Opus 4.5 at $4.50 input and $25 output per million. About a 10x price reduction. If you’re running OpenClas and watching your API bills climb with sub-agents, this is a game-changer. (tho I haven’t tested this myself) Is it the same level of intelligence as whatever magic Anthropic cooks up with Opus? Honestly, I don’t know — there’s something about the Claude models that’s hard to quantify. But for most coding tasks on a budget, you can absolutely switch to Kimi and still get great results.🦞 Clawdbot is no more, Moltbot is dead, Long Live OpenClawAfter we covered the incredible open source project last week, Clawdbot exploded in popularity, driven by Claude Max subscription, and a crazy viral loop where folks who try it, can’t wait to talk about it, it was everywhere! Apparently it was also on Anthropics’ lawyers minds, when they sent Peter Steinberger a friendly worded letter to rebrand and gave him like 12 hours. Apparently, when pronounced, Claude and Clawd sound the same, and they are worried about copyright infringement (which makes sense, most of the early success of Clawd was due to Opus being amazing). The main issue is, due to the popularity of the project, crypto a******s sniped moltybot nickname on X so we got left with Moltbot, which is thematically appropriate, but oh so hard to remember and pronounce!EDIT: OpenClaw was just announced as the new name, apparently I wasn’t the only one who absolutely hated the name Molt! Meanwhile, rebrand or not, my own instance of OpenClaw created an X account, helped me prepare for ThursdAI (including generating a thumbnail), created a video for us today on the fly, and keeps me up to date on emails and unanswered messages via a daily brief. It really has showed me a glimpse of how a truly personal AI assistant can be helpful in a fast changing world! I’ve shared a lot of tips and tricks, about memory, about threads and much more, as we all learn to handle this new ... AI agent framework! But I definitely feel that this is a new unlock in capability, for me and for many others. If you haven’t installed OpenClaw, lmk in the comments why not.Arcee AI Trinity Large: The Western Open Source GiantRemember when we had Lucas Atkins, Arcee’s CTO, on the show just as they were firing up their 2,000 NVIDIA B300 GPUs? Well, the run is complete, and the results are massive. Arcee AI just dropped Trinity Large, a 400B parameter sparse MoE model (with a super efficient 13B active params via 4-of-256 routing) trained on a staggering 17 trillion tokens in just 33 days. This represents the largest publicly announced pretraining run on B300 infrastructure, costing about $20M (and tracked with WandB of course!) and proves that Western labs can still compete at the frontier of open source. Best part? It supports 512K context and is free on OpenRouter until February 2026. Go try it now!Quick open source hits: Trinity Large, Jan v3, DeepSeek OCR updated* Jan AI released Jan v3, a 4B parameter model optimized for local inference. 132 tokens/sec on Apple Silicon, 262K context, 40% improvement on Aider benchmarks. This is the kind of small-but-mighty model you actually can run on your laptop for coding tasks.* Nvidia released PersonaPlex-7B - full duplex voice AI that listens and speaks simultaneously with persona contol* Moonshot AI also releases Kimi Code: Open-source Python-based coding agent with Apache 2.0 licenseVision, Video and AI artxAI Grok Imagine API: #1 in Video GenerationxAI officially launched the Grok Imagine API with an updated model, and it’s now ranked #1 in both text-to-video and image-to-video on the Artificial Analysis leaderboards. It beats Runway Gen-4.5, Kling 2.5 Turbo, and Google Veo 3.1.And of course, the pricing is $4.20 per minute. Of course it is. That’s cheaper than Veo 3.1 at $12/min and Sora 2 Pro at $30/min by 3-7x, with 45-second latency versus 68+ seconds for the competition.During the show, I demoed this live with my AI assistant Wolfred. I literally sent him a message saying “learn this new API based on this URL, take this image of us in the studio, and cre

Hey! Alex here, with another weekly AI update! It seems like ThursdAI is taking a new direction, as this is our 3rd show this year, and a 3rd deep dive into topics (previously Ralph, Agent Skills), please let me know if the comments if you like this format. This week’s deep dive is into Clawdbot, a personal AI assistant you install on your computer, but can control through your phone, has access to your files, is able to write code, help organize your life, but most importantly, it can self improve. Seeing Wolfred (my Clawdbot) learn to transcribe incoming voice messages blew my mind, and I wanted to share this one with you at length! We had Dan Peguine on the show for the deep dive + both Wolfram and Yam are avid users! This one is not to be missed. If ThursdAI is usually too technical for you, use Claude, and install Clawdbot after you read/listen to the deep dive!Also this week, we read Claude’s Constitution that Anthropic released, heard a bunch of new TTS models (some are open source and very impressive) and talked about the new lightspeed coding model GLM 4.7 Flash. First the news, then deep dive, lets go 👇ThursdAI - Recaps of the most high signal AI weekly spaces is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.Open Source AIZ.ai’s GLM‑4.7‑Flash is the Local Agent Sweet Spot (X, HF)This was the open‑source release that mattered this week. Z.ai (formerly Zhipu) shipped GLM‑4.7‑Flash, a 30B MoE model with only 3B active parameters per token, which makes it much more efficient for local agent work. We’re talking a model you can run on consumer hardware that still hits 59% on SWE‑bench Verified, which is uncomfortably close to frontier coding performance. In real terms, it starts to feel like “Sonnet‑level agentic ability, but local.” I know I know, we keep saying “sonnet at home” at different open source models, but this one slaps! Nisten was getting around 120 tokens/sec on an M3 Ultra Mac Studio using MLX, and that’s kind of the headline. The model is fast and capable enough that local agent loops like RALPH suddenly feel practical. It also performs well on browser‑style agent tasks, which is exactly what you want for local automation without sending all your data to a cloud provider. Liquid AI’s LFM2.5‑1.2B Thinking is the “Tiny but Capable” Class (X, HF)Liquid AI released a 1.2B reasoning model that runs under 900MB of memory while still manages to be useful. This thing is built for edge devices and old phones, and the speed numbers are backing it up. We’re talking 239 tok/s decode on AMD CPU, 82 tok/s on mobile NPU, and prefill speeds that make long prompts actually usable. Nisten made a great point: on iOS, there’s a per‑process memory limit around 3.8GB, so a 1.2B model lets you spend your budget on context instead of weights.This is the third class of models we’re now living with: not Claude‑scale, not “local workstation,” but “tiny agent in your pocket.” It’s not going to win big benchmarks, but it’s perfect for on‑device workflows, lightweight assistants, and local RAG.Voice & Audio: Text To Speech is hot this week with 3 releases! We tested three major voice releases this week, and I’m not exaggerating when I say the latency wars are now fully on. Qwen3‑TTS: Open Source, 97ms Latency, Voice Cloning (X, HF)Just 30 minutes before the show, Qwen released their first model of the year, Qwen3 TTS, with two models (0.6B and 1.7B). With support for Voice Cloning based on just 3 seconds of voice, and claims of 97MS latency, this apache 2.0 release looked very good on the surface!The demos we did on stage though... were lackluster. TTS models like Kokoro previously impressed us with super tiny sizes and decent voice, while Qwen3 didn’t really perform on the cloning aspect. For some reason (I tested in Russian which they claim to support) the cloned voice kept repeating the provided sample voice instead of just generating the text I gave it. This confused me, and I’m hoping this is just a demo issue, not a problem with the model. They also support voice design where you just type in the type of voice you want, which to be fair, worked fairly well in our tests!With Apache 2.0 and a full finetuning capability, this is a great release for sure, kudos to the Qwen team! Looking forward to see what folks do with this properly. FlashLabs Chroma 1.0: Real-Time Speech-to-Speech, Open Source (X, HF) Another big open source release in the audio category this week was Chroma 1.0 from FlashLabs, which claim to be the first speech2speech model (not a model that has the traditional ASR>LLM>TTS pipeline) and the claim 150ms end to end latency! The issue with this one is, the company released an open source 4B model, and claimed that this model powers their chat interface demo on the web, but in the release notes they claim the model is english speaking only, while on the website it sounds incredible and I spoke to it in other languages 🤔 I think the mode that we’ve tested is not the open source one. I could’t confirm this at the time of writing, will follow on X with the team and let you guys know. Inworld AI launches TTS-1.5: #1 ranked text-to-speech with sub-250ms latency at half a cent per minute (X, Announcement)Ok this one is definitely in the realm of “voice realistic enough you won’t be able to tell” as this is not an open source model, it’s a new competitor to 11labs and MiniMax - the two leading TTS providers out there. Inworld claims to achieve better results on the TTS Arena, while being significantly cheaper and faster (up to 25x less than leading providers like 11labs) We tested out their voices and they sounded incredible, replied fast and generally was a very good experience. With 130ms response time for their mini version, this is a very decent new entry into the world of TTS providers. Big Companies: Ads in ChatGPT + Claude ConstitutionOpenAI is testing ads in ChatGPT’s free and Go tiers. Ads appear as labeled “Sponsored” content below responses, and OpenAI claim they won’t affect outputs. It’s still a major shift in the product’s business model, and it’s going to shape how people perceive trust in these systems. I don’t love ads, but I understand the economics, they have to make money somehow, with 900M weekly active users, many of them on the free tier, they are bound to make some money with this move. I just hope they won’t turn into a greedy ad optimizing AI machine. Meanwhile, Anthropic released an 80‑page “New Constitution for Claude” that they use during training. This isn’t a prompt, it’s a full set of values baked into the model’s behavior. There’s a fascinating section where they explicitly talk about Claude’s potential wellbeing and how they want to support it. It’s both thoughtful and a little existential. I recommend reading it, especially if you care about alignment and agent design. I applaud Anthropic for releasing this with Creative Commons license for public scrutiny and adoption 👏This weeks buzz - come join the hackathon I’m hosting Jan 31 in SFQuick plug, we have limited seats left open for the hackathon I’m hosting for Weights & Biases at the SF office, and if you’re reading this, and want to join, I’ll approve you if you mention ThursdAI in the application! With sponsors like Redis, Vercel, BrowserBase, Daily, Google Cloud, we are going to give out a LOT of cash as prizes! I’ve also invited a bunch of my friends from the top agentic AI places to be judges, it’s going to be awesome, comeDeep dive into Clawdbot: Local-First, Self-Improving, and Way Too Capable agentClawdbot (C‑L‑A‑W‑D) is that rare project where the hype is justified. It’s an open-source personal agent that runs locally on your Mac, but can talk to you through WhatsApp, Telegram, iMessage, Discord, Slack — basically wherever you already talk. What makes it different is not just the integrations; it’s the self‑improvement loop. You can literally tell it “go build a new skill,” and it will… build the skill, install it, then adopt it and start using it. It’s kind of wild to see it working for the first time. Now... it’s definitely not perfect, far far away from the polish of ChatGPT / Claude, but when it works, damn, it really is mindblowing.That part actually happened live in the episode. Dan Peguine 🐧 showed how he had it create a skill to anonymize his own data so he could demo it on stream without leaking his personal life. Another example: I told my Clawdbot to handle voice notes in Telegram. It didn’t know how, so it went and found a transcription method, wrote itself a skill, saved it, and from that point on just… did the thing. That was the moment it clicked for me. (just before posting this, it forgot how to do it, I think I screwed something up) Dan’s daily brief setup was wild too. It pulls from Apple Health, local calendars, weather, and his own projects, then produces a clean, human daily brief. It also lets him set reminders through WhatsApp and even makes its own decisions about how much to bother him based on context. He shared a moment where it literally told him, “I won’t bug you today because it’s your wife’s birthday.” That isn’t a hardcoded workflow — it’s reasoning layered on top of persistent memory.And that persistent memory is a big deal. It’s stored locally as Markdown files and folders, Obsidian‑style, so you don’t lose your life every time you switch models. You can route the brain to Claude Opus 4.5 today and a local model tomorrow, and the memory stays with you. That is a huge step up from “ChatGPT remembers you unless you unsubscribe.”There’s also a strong community forming around shared skills via ClawdHub. People are building everything from GA4 analytics skills to app testing automations to Tesla battery status checkers. The core pattern is simple but powerful: talk to it, ask it to build a skill, then it can run that skill forever.I definitely have some issues with the security aspect, you are essentially giving

Hey ya’ll, Alex here, and this week I was especially giddy to record the show! Mostly because when a thing clicks for me that hasn’t clicked before, I can’t wait to tell you all about it! This week, that thing is Agent Skills! The currently best way to customize your AI agents with domain expertise, in a simple, repeatable way that doesn’t blow up the context window! We mentioned skills when Anthropic first released them (Oct 16) and when they became an open standard but it didn’t really click until last week! So more on that below. Also this week, Anthropic released a research preview of Claude Cowork, an agentic tool for non coders, OpenAI finally let loos GPT 5.2 Codex (in the API, it was previously available only via Codex), Apple announced a deal with Gemini to power Siri, OpenAI and Anthropic both doubled down on healthcare and much more! We had an incredible show, with an expert in Agent Skills, Eleanor Berger and the usual gang on co-hosts, strongly recommend watching the show in addition to the newsletter! Also, I vibe coded skills support for all LLMs to Chorus, and promised folks a link to download it, so look for that in the footer, let’s dive in! ThursdAI is where you stay up to date! Subscribe to keep us going! Big Company LLMs + APIs: Cowork, Codex, and a Browser in a WeekAnthropic launches Claude Cowork: Agentic AI for Non‑Coders (research preview)Anthropic announced Claude Cowork, which is basically Claude Code wrapped in a friendly UI for people who don’t want to touch a terminal. It’s a research preview available on the Max tier, and it gives Claude read/write access to a folder on your Mac so it can do real work without you caring about diffs, git, or command line.The wild bit is that Cowork was built in a week and a half, and according to the Anthropic team it was 100% written using Claude Code. This feels like a “we’ve crossed a threshold” moment. If you’re wondering why this matters, it’s because coding agents are general agents. If a model can write code to do tasks, it can do taxes, clean your desktop, or orchestrate workflows, and that means non‑developers can now access the same leverage developers have been enjoying for a year.It also isn’t just for files—it comes with a Chrome connector, meaning it can navigate the web to gather info, download receipts, or do research and it uses skills (more on those later)Earlier this week I recorded this first reactions video about Cowork and I’ve been testing it ever since, it’s a very interesting approach of coding agents that “hide the coding” to just... do things. Will this become as big as Claude Code for anthropic (which is reportedly a 1B business for them)? Let’s see! There are real security concerns here, especially if you’re not in the habit of backing up or using git. Cowork sandboxes a folder, but it can still delete things in that folder, so don’t let it loose on your whole drive unless you like chaos.GPT‑5.2 Codex: Long‑Running Agents Are HereOpenAI shipped GPT‑5.2 Codex into the API finally! After being announced as the answer for Opus 4.5 and only being available in Codex. The big headline is SOTA on SWE-Bench and long‑running agentic capability. People describe it as methodical. It takes longer, but it’s reliable on extended tasks, especially when you let it run without micromanaging.This model is now integrated into Cursor, GitHub Copilot, VS Code, Factory, and Vercel AI Gateway within hours of launch. It’s also state‑of‑the‑art on SWE‑Bench Pro and Terminal‑Bench 2.0, and it has native context compaction. That last part matters because if you’ve ever run an agent for long sessions, the context gets bloated and the model gets dumber. Compaction is an attempt to keep it coherent by summarizing old context into fresh threads, and we debated whether it really works. I think it helps, but I also agree that the best strategy is still to run smaller, atomic tasks with clean context.Cursor vibe-coded browser with GPT-5.2 and 3M lines of codeThe most mind‑blowing thing we discussed is Cursor letting GPT‑5.2 Codex run for a full week to build a browser called FastRenderer. This is not Chromium‑based. It’s a custom HTML parser, CSS cascade, layout engine, text shaping, paint pipeline, and even a JavaScript VM, written in Rust, from scratch. The codebase is open source on GitHub, and the full story is on Cursor’s blog It took nearly 30,000 commits and millions of lines of code. The system ran hundreds of concurrent agents with a planner‑worker architecture, and GPT‑5.2 was the best model for staying on task in that long‑running regime. That’s the real story, not just “lol a model wrote a browser.” This is a stress test for long‑horizon agentic software development, and it’s a preview of how teams will ship in 2026.I said on the show, browsers are REALLY hard, it took two decades for the industry to settle and be able to render websites normally, and there’s a reason everyone’s using Chromium. This is VERY impressive 👏 Now as for me, I began using Codex again, but I still find Opus better? Not sure if this is just me expecting something that’s not there? I’ll keep you postedGemini Personal Intelligence: The Data Moat king is back! What kind of car do you drive? Does ChatGPT know that? welp, it turns our Google does (based on your emails, Google photos) and now Gemini can tap into this personal info (if you allow it, they are stressing privacy), and give you much more personalized answers! Flipping this Beta feature on, lets Gemini reason across Gmail, YouTube, Photos, and Search with explicit opt‑in permissions, and it’s rolling out to Pro and Ultra users in the US first.I got to try it early, and it’s uncanny. I asked Gemini what car I drive, and it told me I likely drive a Model Y, but it noticed I recently searched for a Honda Odyssey and asked if I was thinking about switching. It was kinda... freaky because I forgot I had early access and this was turned on 😂 Pro Tip: if you’re brave enough to turn this on, ask for a complete profile on you 🙂Now the last piece is for Gemini to become proactive, suggesting things for me based on my needs! Apple & Google: The Partnership (and Drama Corner)We touched on this in the intro, but it’s official: Apple Intelligence will be powered by Google Gemini for “world knowledge” tasks. Apple stated that after “careful evaluation,” Google provided the most capable foundation model for their.. apple foundation models. It’s confusing, I agree.Honestly? I got excited about Apple Intelligence, but Siri is still... Siri. It’s 2026 and we are still struggling with basic intents. Hopefully, plugging Gemini into the backend changes that? In other drama: The silicon valley carousel continues. 3 Co-founders (Barret Zoph, Sam Schoenholz and Luke Metz) from Thinking Machines (and former OpenAI folks) have returned to the mothership (OpenAI), amid some vague tweets about “unethical conduct.” It’s never a dull week on the timeline. This Week’s Buzz: WeaveHacks 3 in SFI’ve got one thing in the Buzz corner this week, and it’s a big one. WeaveHacks 3 is back in San Francisco, January 31st - February 1st. The theme is self‑improving agents, and if you’ve been itching to build in person, this is it. We’ve got an amazing judge lineup, incredible sponsors, and a ridiculous amount of agent tooling to play with.You can sign up here: https://luma.com/weavehacks3If you’re coming, add to the form you heard it on ThursdAI and we’ll make sure you get in! Deep Dive: Agent Skills With Eleanor BergerThis was the core of the episode, and I’m still buzzing about it. We brought on Eleanor Berger, who has basically become the skill evangelist for the entire community, and she walked us through why skills are the missing layer in agentic AI.Skills are simple markdown files with a tiny bit of metadata in a directory together optional scripts, references, and assets. The key idea is progressive disclosure. Instead of stuffing your entire knowledge base into the context, the model only sees a small list of skills and let it load only what it needs. That means you can have hundreds of skills without blowing your context window (and making the model dumber and slower in result) The technical structure is dead simple, but the implications are huge. Skills create a portable, reusable, composable way to give agents domain expertise, and they now work across most major harnesses. That means you can build a skill once and use it in Claude, Cursor, AMP, or any other agent tool that supports the standard.Eleanor made the point that skills are an admission that we now have general‑purpose agents. The model can do the work, but it doesn’t know your preferences, your domain, your workflows. Skills are how you teach it those things. We also talked about how scripts inside skills reduce variance because you’re not asking the model to invent code every time; you’re just invoking trusted tools.What really clicked for me this week is how easy it is to create skills using an agent. You don’t need to hand‑craft directories. You can describe your workflow, or even just do the task once in chat, and then ask the agent to turn it into a skill. It really is very very simple! And that’s likely the reason everyone is adopting this simple formart for extension their agents knowledge.Get started with skillsIf you use Claude Chat, the simplest way to get started is ask Claude to review your previous conversations and suggest a skill for you. Or, at the end of a long chat where you went back and forth with Claude on a task, ask it to distill the important parts into a skill. If you want to use other people’s skills, and you are using Claude Code, or any of the supported IDE/Agents, here’s where to download the folders and install them: If you aren’t a developer and don’t subscribe to Claude, well, I got good news for you! I vibecoded skill support for every LLM 👇The Skills Demo That Changed My MindI was resistant to skills at first, mostly because I wanted them insi

Hey folks, Alex here from Weights & Biases, with your weekly AI update (and a first live show of this year!) For the first time, we had a co-host of the show also be a guest on the show, Ryan Carson (from Amp) went supernova viral this week with an X article (1.5M views) about Ralph Wiggum (yeah, from Simpsons) and he broke down that agentic coding technique at the end of the show. LDJ and Nisten helped cover NVIDIA’s incredible announcements during CES with their Vera Rubin upcoming platform (4-5X improvements) and we all got excited about AI medicine with ChatGPT going into Health officially! Plus, a bunch of Open Source news, let’s get into this: ThursdAI - Recaps of the most high signal AI weekly spaces is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.Open Source: The “Small” Models Are WinningWe often talk about the massive frontier models, but this week, Open Source came largely from unexpected places and focused on efficiency, agents, and specific domains.Solar Open 100B: A Data MasterclassUpstage released Solar Open 100B, and it’s a beast. It’s a 102B parameter Mixture-of-Experts (MoE) model, but thanks to MoE magic, it only uses about 12B active parameters during inference. This means it punches incredibly high but runs fast.What I really appreciated here wasn’t just the weights, but the transparency. They released a technical report detailing their “Data Factory” approach. They trained on nearly 20 trillion tokens, with a huge chunk being synthetic. They also used a dynamic curriculum that adjusted the difficulty and the ratio of synthetic data as training progressed. This transparency is what pushes the whole open source community forward.Technically, it hits 88.2 on MMLU and competes with top-tier models, especially in Korean language tasks. You can grab it on Hugging Face.MiroThinker 1.5: The DeepSeek Moment for Agents?We also saw MiroThinker 1.5, a 30B parameter model that is challenging the notion that you need massive scale to be smart. It uses something they call “Interactive Scaling.”Wolfram broke this down for us: this agent forms hypotheses, searches for evidence, and then iteratively revises its answers in a time-sensitive sandbox. It effectively “thinks” before answering. The result? It beats trillion-parameter models on search benchmarks like BrowseComp. It’s significantly cheaper to run, too. This feels like the year where smaller models + clever harnesses (harnesses are the software wrapping the model) will outperform raw scale.Liquid AI LFM 2.5: Running on Toasters (Almost)We love Liquid AI and they are great friends of the show. They announced LFM 2.5 at CES with AMD, and these are tiny ~1B parameter models designed to run on-device. We’re talking about running capable AI on your laptop, your phone, or edge devices (or the Reachy Mini bot that I showed off during the show! I gotta try and run LFM on him!)Probably the coolest part is the audio model. Usually, talking to an AI involves a pipeline: Speech-to-Text (ASR) -> LLM -> Text-to-Speech (TTS). Liquid’s model is end-to-end. It hears audio and speaks audio directly. We watched a demo from Maxime Labonne where the model was doing real-time interaction, interleaving text and audio. It’s incredibly fast and efficient. While it might not write a symphony for you, for on-device tasks like summarization or quick interactions, this is the future.NousCoder-14B and Zhipu AI IPOA quick shoutout to our friends at Nous Research who released NousCoder-14B, an open-source competitive programming model that achieved a 7% jump on LiveCodeBench accuracy in just four days of RL training on 48 NVIDIA B200 GPUs. The model was trained on 24,000 verifiable problems, and the lead researcher Joe Li noted it achieved in 4 days what took him 2 years as a teenager competing in programming contests. The full RL stack is open-sourced on GitHub and Nous published a great WandB results page as well! And in historic news, Zhipu AI (Z.ai)—the folks behind the GLM series—became the world’s first major LLM company to IPO, raising $558 million on the Hong Kong Stock Exchange. Their GLM-4.7 currently ranks #1 among open-source and domestic models on both Artificial Analysis and LM Arena. Congrats to them!Big Companies & APIsNVIDIA CES: Vera Rubin Changes EverythingLDJ brought the heat on this one covering Jensen’s CES keynote that unveiled the Vera Rubin platform, and the numbers are almost hard to believe. We’re talking about a complete redesign of six chips: the Rubin GPU delivering 50 petaFLOPS of AI inference (5x Blackwell), the Vera CPU with 88 custom Olympus ARM cores, NVLink 6, ConnectX-9 SuperNIC, BlueField-4 DPU, and Spectrum-6 Ethernet.Let me put this in perspective using LDJ’s breakdown: if you look at FP8 performance, the jump from Hopper to Blackwell was about 5x. The jump from Blackwell to Vera Rubin is over 3x again—but here’s the kicker—while only adding about 200 watts of power draw. That’s insane efficiency improvement.The real-world implications Jensen shared: training a 10 trillion parameter mixture-of-experts model now requires 75% fewer GPUs compared to Blackwell. Inference token costs drop roughly 10x—a 1MW cluster goes from 1 million to 10 million tokens per second at the same power. HBM4 memory delivers 22 TB/s bandwidth with 288GB capacity, exceeding NVIDIA’s own 2024 projections by nearly 70%.As Ryan noted, when people say there’s an AI bubble, this is why it’s hilarious. Jensen keeps saying the need for inference is unbelievable and only going up exponentially. We all see this. I can’t get enough inference—I want to spin up 10 Ralphs running concurrently! The NVL72 rack-scale system achieves 3.6 exaFLOPS inference with 20.7TB total HBM, and it’s already shipping. Runway 4.5 is already running on the new platform, having ported their model from Hopper to Vera Rubin NVL72 in a single day.NVIDIA also recently acqui-hidred Groq (with a Q) in a ~$20 billion deal, bringing the inference chip expertise from the guy who created Google’s TPUs in-house.Nemotron Speech ASR & The Speed of Voice (X, HF, Blog)NVIDIA also dropped Nemotron Speech ASR. This is a 600M parameter model that offers streaming transcription with 24ms latency.We showed a demo from our friend Kwindla Kramer at Daily. He was talking to an AI, and the response was virtually instant. The pipeline is: Nemotron (hearing) -> Llama/Nemotron Nano (thinking) -> Magpie TTS (speaking). The total latency is under 500ms. It feels like magic. Instant voice agents are going to be everywhere this year.XAI Raises $20B While Grok Causes Problems (Again)So here’s the thing about covering anything Elon-related: it’s impossible to separate signal from noise because there’s an army of fans who hype everything and an army of critics who hate everything. But let me try to be objective here.XAI raised another massive Round E of $20 billion! at a $230 billion valuation, with NVIDIA and Cisco as strategic investors. The speed of their infrastructure buildout is genuinely incredible. Grok’s voice mode is impressive. I use Grok for research and it’s really good, notable for it’s unprecedented access to X !But. This raise happened in the middle of a controversy where Grok’s image model was being used to “put bikinis” on anyone in reply threads, including—and this is where I draw a hard line—minors. As Nisten pointed out on the show, it’s not even hard to implement guardrails. You just put a 2B VL model in front and ask “is there a minor in this picture?” But people tested it, asked Grok not to use the feature, and it did it anyway. And yeah, putting Bikini on Claude is funny, but basic moderation is lacking! The response of “we’ll prosecute illegal users” is stupid when there’s no moderation built into the product. There’s an enormous difference between Photoshop technically being able to do something after hours of work, and a feature that generates edited images in one second as the first comment to a celebrity, then gets amplified by the platform’s algorithm to millions of people. One is a tool. The other is a product with amplification mechanics. Products need guardrails. I don’t often link to CNN (in fact this is the first time) but they have a great writeup about the whole incident here which apparently includes the quitting of a few trust and safety folks and Elon’s pushback on guardrails. CrazyThat said, Grok 5 is in training and XAI continues to ship impressive technology. I just wish they’d put the same engineering effort into safety as they do into capabilities!OpenAI Launches GPT HealthThis one’s exciting. OpenAI CEO Fidji Simo announced ChatGPT Health, a privacy-first space for personalized health conversations that can connect to electronic health records, Apple Health, Function Health, Peloton, and MyFitnessPal.Here’s why this matters: health already represents about 5% of all ChatGPT messages globally and touches 25% of weekly active users—often outside clinic hours or in underserved areas. People are already using these models for health advice constantly.Nisten, who has worked on AI doctors since the GPT-3 days and even published papers on on-device medical AI, gave us some perspective: the models have been fantastic for health stuff for two years now. The key insight is that medical data seems like a lot, but there are really only about 2,000 prescription drugs and 2,000 diseases (10,000 if you count rare ones). That’s nothing for an LLM. The models excel at pattern recognition across this relatively contained dataset.The integration with Function Health is particularly interesting to me. Function does 160+ lab tests, but many doctors won’t interpret them because they didn’t order them. ChatGPT could help bridge that gap, telling you “hey, this biomarker looks off, you should discuss this with your doctor.” The bad news is, this is just a waitlist and you can add yourself to the waitlist here, we’ll keep monitoring the

Hey all, Happy new year! This is Alex, writing to you for the very fresh start of this year, it’s 2026 already, can you believe it? There was no live stream today, I figured the cohosts deserve a break and honestly it was a very slow week. Even the chinese labs who don’t really celebrate X-mas and new years didn’t come out with a banger AFAIK. ThursdAI - AI moves fast, we’re here to make sure you never miss a thing! Subscribe :) Tho I thought it was an incredible opportunity to finally post the Will Brow interview I recorded in November during the AI Engineer conference. Will is a researcher at Prime Intellect (big fans on WandB btw!) and is very known on X as a hot takes ML person, often going viral for tons of memes! Will is the creator and maintainer of the Verifiers library (Github) and his talk at AI Engineer was all about RL Environments (what they are, you can hear in the interview, I asked him!) TL;DR last week of 2025 in AIBesides this, my job here is to keep you up to date, and honestly this was very easy this week, as… almost nothing has happened, but here we go: Meta buys ManusThe year ended with 2 huge acquisitions / aquihires. First we got the news from Alex Wang that Meta has bought Manus.ai which is an agentic AI startup we covered back in March for an undisclosed amount (folks claim $2-3B) The most interesting thing here is that Manus is a Chinese company, and this deal requires very specific severance from Chinese operations.Jensen goes on a new years spending spree, Nvidia buys Groq (not GROK) for $20BGroq which we covered often here, and are great friends, is going to NVIDIA, in a… very interesting acqui-hire, which is a “non binding license” + most of Groq top employees apparently are going to NVIDIA. Jonathan Ross the CEO of Groq, was the co-creator of the TPU chips at Google before founding Groq, so this seems like a very strategic aquihire for NVIDIA! Congrats to our friends from Groq on this amazing news for the new year! Tencent open-sources HY-MT1.5 translation models with 1.8B edge-deployable and 7B cloud variants supporting 33 languages (X, HF, HF, GitHub)It seems that everyone’s is trying to de-throne whisper and this latest attempt from Tencent is a interesting one. a 1.8B and 7B translation models with very interesting stats. Alibaba’s Qwen-Image-2512 drops on New Year’s Eve as strongest open-source text-to-image model, topping AI Arena with photorealistic humans and sharper textures (X, HF, Arxiv)Our friends in Tongyi decided to give is a new years present in the form of an updated Qwen-image, with much improved realismThat’s it folks, this was a quick one, hopefully you all had an amazing new year celebration, and are gearing up to an eventful and crazy 2026. I wish you all happiness, excitement and energy to keep up with everything in the new year, and will make sure that we’re here to keep you up to date as always! P.S - I got a little news of my own this yesterday, not related to AI. She said yes 🎉 This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit sub.thursdai.news/subscribe

Ho Ho Ho, Alex here! (a real human writing these words, this needs to be said in 2025) Merry Christmas (to those who celebrate) and welcome to the very special yearly ThursdAI recap! This was an intense year in the world of AI, and after 51 weekly episodes (this is episode 52!) we have the ultimate record of all the major and most important AI releases of this year! So instead of bringing you a weekly update (it’s been a slow week so far, most AI labs are taking a well deserved break, the Cchinese AI labs haven’t yet surprised anyone), I’m dropping a comprehensive yearly AI review! Quarter by quarter, month by month, both in written form and as a pod/video! Why do this? Who even needs this? Isn’t most of it obsolete? I have asked myself this exact question while prepping for the show (it was quite a lot of prep, even with Opus’s help). I eventually landed on, hey, if nothing else, this will serve as a record of the insane week of AI progress we all witnessed. Can you imagine that the term Vibe Coding is less than 1 year old? That Claude Code was released at the start of THIS year? We get hedonicly adapt to new AI goodies so quick, and I figured this will serve as a point in time check, we can get back to and feel the acceleration! With that, let’s dive in - P.S. the content below is mostly authored by my co-author for this, Opus 4.5 high, which at the end of 2025 I find the best creative writer with the best long context coherence that can imitate my voice and tone (hey, I’m also on a break! 🎅) “Open source AI has never been as hot as this quarter. We’re accelerating as f*ck, and it’s only just beginning—hold on to your butts.” — Alex Volkov, ThursdAI Q1 2025🏆 The Big Picture — 2025 - The Year the AI Agents Became RealLooking back at 51 episodes and 12 months of relentless AI progress, several mega-themes emerged:1. 🧠 Reasoning Models Changed EverythingFrom DeepSeek R1 in January to GPT-5.2 in December, reasoning became the defining capability. Models now think for hours, call tools mid-thought, and score perfect on math olympiads.2. 🤖 2025 Was Actually the Year of AgentsWe said it in January, and it came true. Claude Code launched the CLI revolution, MCP became the universal protocol, and by December we had ChatGPT Apps, Atlas browser, and AgentKit.3. 🇨🇳 Chinese Labs Dominated Open SourceDeepSeek, Qwen, MiniMax, Kimi, ByteDance — despite chip restrictions, Chinese labs released the best open weights models all year. Qwen 3, Kimi K2, DeepSeek V3.2 were defining releases.4. 🎬 We Crossed the Uncanny ValleyVEO3’s native audio, Suno V5’s indistinguishable music, Sora 2’s social platform — 2025 was the year AI-generated media became indistinguishable from human-created content.5. 💰 The Investment Scale Became Absurd$500B Stargate, $1.4T compute obligations, $183B valuations, $100-300M researcher packages, LLMs training in space. The numbers stopped making sense.6. 🏆 Google Made a ComebackAfter years of “catching up,” Google delivered Gemini 3, Antigravity, Nano Banana Pro, VEO3, and took the #1 spot (briefly). Don’t bet against Google.By the NumbersQ1 2025 — The Quarter That Changed EverythingDeepSeek R1 crashed NVIDIA’s stock, reasoning models went mainstream, and Chinese labs took over open source. The quarter that proved AI isn’t slowing down—it’s just getting started.Key Themes:* 🧠 Reasoning models went mainstream (DeepSeek R1, o1, QwQ)* 🇨🇳 Chinese labs dominated open source (DeepSeek, Alibaba, MiniMax, ByteDance)* 🤖 2025 declared “The Year of Agents” (OpenAI Operator, MCP won)* 🖼️ Image generation revolution (GPT-4o native image gen, Ghibli-mania)* 💰 Massive infrastructure investment (Project Stargate $500B)January — DeepSeek Shakes the World(Jan 02 | Jan 10 | Jan 17 | Jan 24 | Jan 30)The earthquake that shattered the AI bubble. DeepSeek R1 dropped on January 23rd and became the most impactful open source release ever:* Crashed NVIDIA stock 17% — $560B loss, largest single-company monetary loss in history* Hit #1 on the iOS App Store* Cost allegedly only $5.5M to train (sparking massive debate)* Matched OpenAI’s o1 on reasoning benchmarks at 50x cheaper pricing* The 1.5B model beat GPT-4o and Claude 3.5 Sonnet on math benchmarks 🤯“My mom knows about DeepSeek—your grandma probably knows about it, too” — Alex VolkovAlso this month:* OpenAI Operator — First agentic ChatGPT (browser control, booking, ordering)* Project Stargate — $500B AI infrastructure (Manhattan Project for AI)* NVIDIA Project Digits — $3,000 desktop that runs 200B parameter models* Kokoro TTS — 82M param model hit #1 on TTS Arena, Apache 2, runs in browser* MiniMax-01 — 4M context window from Hailuo* Gemini Flash Thinking — 1M token context with thinking tracesFebruary — Reasoning Mania & The Birth of Vibe Coding(Feb 07 | Feb 13 | Feb 20 | Feb 28)The month that redefined how we work with AI.OpenAI Deep Research (Feb 6) — An agentic research tool that scored 26.6% on Humanity’s Last Exam (vs 10% for o1/R1). Dr. Derya Unutmaz called it “a phenomenal 25-page patent application that would’ve cost $10,000+.”Claude 3.7 Sonnet & Claude Code (Feb 24-27) — Anthropic’s coding beast hit 70% on SWE-Bench with 8x more output (64K tokens). Claude Code launched as Anthropic’s agentic coding tool — marking the start of the CLI agent revolution.“Claude Code is just exactly in the right stack, right around the right location... You can do anything you want with a computer through the terminal.” — Yam PelegGPT-4.5 (Orion) (Feb 27) — OpenAI’s largest model ever (rumored 10T+ parameters). 62.5% on SimpleQA, foundation for future reasoning models.Grok 3 (Feb 20) — xAI enters the arena with 1M token context and “free until GPUs melt.”Andrej Karpathy coins “Vibe Coding” (Feb 2) — The 5.2M view tweet that captured a paradigm shift: developers describe what they want, AI handles implementation.OpenAI Roadmap Revelation (Feb 13) — Sam Altman announced GPT-4.5 will be the last non-chain-of-thought model. GPT-5 will unify everything.March — Google’s Revenge & The Ghibli Explosion(Mar 06 | Mar 13 | Mar 20 | Mar 27)Gemini 2.5 Pro Takes #1 (Mar 27) — Google reclaimed the LLM crown with AIME jumping nearly 20 points, 1M context, “thinking” integrated into the core model.GPT-4o Native Image Gen — Ghibli-mania (Mar 27) — The internet lost its collective mind and turned everything into Studio Ghibli. Auto-regressive image gen with perfect text rendering, incredible prompt adherence.“The internet lost its collective mind and turned everything into Studio Ghibli” — Alex VolkovMCP Won (Mar 27) — OpenAI officially adopted Anthropic’s Model Context Protocol. No VHS vs Betamax situation. Tools work across Claude AND GPT.DeepSeek V3 685B — AIME jumped from 39.6% → 59.4%, MIT licensed, best non-reasoning open model.ThursdAI Turns 2! (Mar 13) — Two years since the first episode about GPT-4.Open Source Highlights:* Gemma 3 (1B-27B) — 128K context, multimodal, 140+ languages, single GPU* QwQ-32B — Qwen’s reasoning model matches R1, runs on Mac* Mistral Small 3.1 — 24B, beats Gemma 3, Apache 2* Qwen2.5-Omni-7B — End-to-end multimodal with speech outputQ2 2025 — The Quarter That Shattered RealityVEO3 crossed the uncanny valley, Claude 4 arrived with 80% SWE-bench, and Qwen 3 proved open source can match frontier models. The quarter we stopped being able to tell what’s real.Key Themes:* 🎬 Video AI crossed the uncanny valley (VEO3 with native audio)* 🧠 Tool-using reasoning models emerged (o3 calling tools mid-thought)* 🇨🇳 Open source matched frontier (Qwen 3, Claude 4)* 📺 Google I/O delivered everything* 💸 AI’s economic impact accelerated ($300B valuations, 80% price drops)April — Tool-Using Reasoners & Llama Chaos(Apr 03 | Apr 10 | Apr 17 | Apr 24)OpenAI o3 & o4-mini (Apr 17) — The most important reasoning upgrade ever. For the first time, o-series models can use tools during reasoning: web search, Python, image gen. Chain 600+ consecutive tool calls. Manipulate images mid-thought.“This is almost AGI territory — agents that reason while wielding tools” — Alex VolkovGPT-4.1 Family (Apr 14) — 1 million token context across all models. Near-perfect recall. GPT-4.5 deprecated.Meta Llama 4 (Apr 5) — Scout (17B active/109B total) & Maverick (17B active/400B total). LMArena drama (tested model ≠ released model). Community criticism. Behemoth teased but never released.Gemini 2.5 Flash (Apr 17) — Set “thinking budget” per API call. Ultra-cheap at $0.15/$0.60 per 1M tokens.ThursdAI 100th Episode! 🎉May — VEO3 Crosses the Uncanny Valley & Claude 4 Arrives(May 01 | May 09 | May 16 | May 23 | May 29)VEO3 — The Undisputed Star of Google I/O (May 20) — Native multimodal audio generation (speech, SFX, music synced perfectly). Perfect lip-sync. Characters understand who’s speaking. Spawned viral “Prompt Theory” phenomenon.“VEO3 isn’t just video generation — it’s a world simulator. We crossed the uncanny valley this quarter.” — Alex VolkovClaude 4 Opus & Sonnet — Live Drop During ThursdAI! (May 22) — Anthropic crashed the party mid-show. First models to cross 80% on SWE-bench. Handles 6-7 hour human tasks. Hybrid reasoning + instant response modes.Qwen 3 (May 1) — The most comprehensive open source release ever: 8 models, Apache 2.0. Runtime /think toggle for chain-of-thought. 4B dense beats Qwen 2.5-72B on multiple benchmarks. 36T training tokens, 119 languages.“The 30B MoE is ‘Sonnet 3.5 at home’ — 100+ tokens/sec on MacBooks” — NistenGoogle I/O Avalanche:* Gemini 2.5 Pro Deep Think (84% MMMU)* Jules (free async coding agent)* Project Mariner (browser control via API)* Gemini Ultra tier ($250/mo)June — The New Normal(Jun 06 | Jun 13 | Jun 20 | Jun 26)o3 Price Drop 90% (Jun 12) — From $40/$10 → $8/$2 per million tokens. o3-pro launched at 87% cheaper than o1-pro.Meta’s $15B Scale AI Power Play (Jun 12) — 49% stake in Scale AI. Alex Wang leads new “Superintelligence team” at Meta. Seven-to-nine-figure comp packages for researchers.MiniMax M1 — Reasoning MoE That Beats R1 (Jun 19) — 456B total / 45B active par

Hey folks 👋 Alex here, dressed as 🎅 for our pre X-mas episode!We’re wrapping up 2025, and the AI labs decided they absolutely could NOT let the year end quietly. This week was an absolute banger—we had Gemini 3 Flash dropping with frontier intelligence at flash prices, OpenAI firing off GPT 5.2 Codex as breaking news DURING our show, ChatGPT Images 1.5, Nvidia going all-in on open source with Nemotron 3 Nano, and the voice AI space heating up with Grok Voice and Chatterbox Turbo. Oh, and Google dropped FunctionGemma for all your toaster-to-fridge communication needs (yes, really).Today’s show was over three and a half hours long because we tried to cover both this week AND the entire year of 2025 (that yearly recap is coming next week—it’s a banger, we went month by month and you’ll really feel the acceleration). For now, let’s dive into just the insanity that was THIS week.00:00 Introduction and Overview00:39 Weekly AI News Highlights01:40 Open Source AI Developments01:44 Nvidia's Nemotron Series09:09 Google's Gemini 3 Flash19:26 OpenAI's GPT Image 1.520:33 Infographic and GPT Image 1.5 Discussion20:53 Nano Banana vs GPT Image 1.521:23 Testing and Comparisons of Image Models23:39 Voice and Audio Innovations24:22 Grok Voice and Tesla Integration26:01 Open Source Robotics and Voice Agents29:44 Meta's SAM Audio Release32:14 Breaking News: Google Function Gemma33:23 Weights & Biases Announcement35:19 Breaking News: OpenAI Codex 5.2 MaxTo receive new posts and support my work, consider becoming a free or paid subscriber.Big Companies LLM updatesGoogle’s Gemini 3 Flash: The High-Speed Intelligence KingIf we had to title 2025, as Ryan Carson mentioned on the show, it might just be “The Year of Google’s Comeback.” Remember at the start of the year when we were asking “Where is Google?” Well, they are here. Everywhere.This week they launched Gemini 3 Flash, and it is rightfully turning heads. This is a frontier-class model—meaning it boasts Pro-level intelligence—but it runs at Flash-level speeds and, most importantly, Flash-level pricing. We are talking $0.50 per 1 million input tokens. That is not a typo. The price-to-intelligence ratio here is simply off the charts.I’ve been using Gemini 2.5 Flash in production for a while because it was good enough, but Gemini 3 Flash is a different beast. It scores 71 on the Artificial Analysis Intelligence Index (a 13-point jump from the previous Flash), and it achieves 78% on SWE-bench Verified. That actually beats the bigger Gemini 3 Pro on some agentic coding tasks!What impressed me most, and something Kwindla pointed out, is the tool calling. Previous Gemini models sometimes struggled with complex tool use compared to OpenAI, but Gemini 3 Flash can handle up to 100 simultaneous function calls. It’s fast, it’s smart, and it’s integrated immediately across the entire Google stack—Workspace, Android, Chrome. Google isn’t just releasing models anymore; they are deploying them instantly to billions of users.For anyone building agents, this combination of speed, low latency, and 1 million context window (at this price!) makes it the new default workhorse.Google’s FunctionGemma Open Source releaseWe also got a smaller, quirkier release from Google: FunctionGemma. This is a tiny 270M parameter model. Yes, millions, not billions.It’s purpose-built for function calling on edge devices. It requires only 500MB of RAM, meaning it can run on your phone, in your browser, or even on a Raspberry Pi. As Nisten joked on the show, this is finally the model that lets your toaster talk to your fridge.Is it going to write a novel? No. But after fine-tuning, it jumped from 58% to 85% accuracy on mobile action tasks. This represents a future where privacy-first agents live entirely on your device, handling your calendar and apps without ever pinging a cloud server.OpenAI Image 1.5, GPT 5.2 Codex and ChatGPT AppstoreOpenAI had a busy week, starting with the release of GPT Image 1.5. It’s available now in ChatGPT and the API. The headline here is speed and control—it’s 4x faster than the previous model and 20% cheaper. It also tops the LMSYS Image Arena leaderboards.However, I have to give a balanced take here. We’ve been spoiled recently by Google’s “Nano Banana Pro” image generation (which powers Gemini). When we looked at side-by-side comparisons, especially with typography and infographic generation, Gemini often looked sharper and more coherent. This is what we call “hedonistic adaptation”—GPT Image 1.5 is great, but the bar has moved so fast that it doesn’t feel like the quantum leap DALL-E 3 was back in the day. Still, for production workflows where you need to edit specific parts of an image without ruining the rest, this is a massive upgrade.🚨 BREAKING: GPT 5.2 CodexJust as we were nearing the end of the show, OpenAI decided to drop some breaking news: GPT 5.2 Codex.This is a specialized model optimized specifically for agentic coding, terminal workflows, and cybersecurity. We quickly pulled up the benchmarks live, and they look significant. It hits 56.4% on SWE-Bench Pro and a massive 64% on Terminal-Bench 2.0.It supports up to 400k token inputs with native context compaction, meaning it’s designed for those long, complex coding sessions where you’re debugging an entire repository. The coolest (and scariest?) stat: a security researcher used this model to find three previously unknown vulnerabilities in React in just one week.OpenAI is positioning this for “professional software engineering,” and the benchmarks suggest a 30% improvement in token efficiency over the standard GPT 5.2. We are definitely going to be putting this through its paces in our own evaluations soon.ChatGPT ... the AppStore!Also today (OpenAI is really throwing everything they have to the end of the year release party), OpenAI has unveiled how their App Store is going to look and opened the submission forms to submit your own apps!Reminder, ChatGPT apps are powered by MCP and were announced during DevDay, they let companies build a full UI experience right inside ChatGPT, and given OpenAi’s almost 900M weekly active users, this is a big deal! Do you have an app you’d like in there? let me know in the comments!Open Source AI🔥 Nvidia Nemotron 3 Nano: The Most Important Open Source Release of the Week (X, HF)I think the most important release of this week in open source was Nvidia Nemotron 3 Nano, and it was pretty much everywhere. Nemotron is a series of models from Nvidia that’s been pushing efficiency updates, finetune innovations, pruning, and distillations—all the stuff Nvidia does incredibly well.Nemotron 3 Nano is a 30 billion parameter model with only 3 billion active parameters, using a hybrid Mamba-MoE architecture. This is huge. The model achieves 1.5 to 3.3x faster inference than competing models like Qwen 3 while maintaining competitive accuracy on H200 GPUs.But the specs aren’t even the most exciting part. NVIDIA didn’t just dump the weights over the wall. They released the datasets—all 25 trillion tokens of pre-training and post-training data. They released the recipes. They released the technical reports. This is what “Open AI” should actually look like.What’s next? Nemotron 3 Super at 120B parameters (4x Nano) and Nemotron 3 Ultra at 480B parameters (16x Nano) are coming in the next few months, featuring their innovative Latent Mixture of Experts architecture.Check out the release on HuggingFaceOther Open Source HighlightsLDJ brought up BOLMO from Allen AI—the first byte-level model that actually reaches parity with similar-size models using regular tokenization. This is really exciting because it could open up new possibilities for spelling accuracy, precise code editing, and potentially better omnimodality since ultimately everything is bytes—images, audio, everything.Wolfram highlighted OLMO 3.1, also from Allen AI, which is multimodal with video input in three sizes (4B, 7B, 8B). The interesting feature here is that you can give it a video, ask something like “how many times does a ball hit the crown?” and it’ll not only give you the answer but mark the precise coordinates on the video frames where it happens. Very cool for tracking objects throughout a video!Mistral OCR 3 (X)Mistral also dropped Mistral OCR 3 this week—their next-generation document intelligence model achieving a 74% win rate over OCR 2 across challenging document types. We’re talking forms, low-quality scans, handwritten text, complex tables, and multilingual documents.The pricing is aggressive at just $2 per 1,000 pages (or $1 with Batch API discount), and it outperforms enterprise solutions like AWS Textract, Azure Doc AI, and Google DocSeek. Available via API and their new Document AI Playground.🐝 This Week’s Buzz: Wolfram Joins Weights & Biases!I am so, so hyped to announce this. Our very own co-host and evaluation wizard, Wolfram RavenWlf, is officially joining the Weights & Biases / CoreWeave family as an AI Evangelist and “AIvaluator” starting in January!Wolfram has been the backbone of the “vibe checks” and deep-dive evals on this show for a long time. Now, he’ll be doing it full-time, building out benchmarks for the community and helping all of us make sense of this flood of models. Expect ThursdAI to get even more data-driven in 2026. Match made in heaven! And if you’re as excited as we are, give Weave a try, it’s free to get started!Voice & Audio: Faster, Cheaper, BetterIf 2025 was the year of the LLM comeback, the end of 2025 is the era of Voice AI commoditization. It is getting so cheap and so fast.Grok Voice Agent API (X)xAI launched their Grok Voice Agent API, and the pricing is aggressive: $0.05 per minute flat rate. That significantly undercuts OpenAI and others. But the real killer feature here is the integration.If you drive a Tesla, this is what powers the voice command when you hold down the button. It has native access to vehicle controls, but for developers, it has native tool calling for Real-time X Search. Th

Hey everyone, December started strong and does NOT want to slow down!? OpenAI showed us their response to the Code Red and it’s GPT 5.2, which doesn’t feel like a .1 upgrade! We got it literally as breaking news at the end of the show, and oh boy! The new kind of LLMs is here. GPT, then Gemini, then Opus and now GPT again... Who else feels like we’re on a trippy AI rolercoaster? Just me? 🫨 I’m writing this newsletter from a fresh “traveling podcaster” setup in SF (huge shoutout to the Chroma team for the studio hospitality). P.S - Next week we’re doing a year recap episode (52st episode of the year, what is my life), but today is about the highest-signal stuff that happened this week.Alright. No more foreplay. Let’s dive in. Please subscribe. 🔥 The main event: OpenAI launches GPT‑5.2 (and it’s… a lot)We started the episode with “garlic in the air” rumors (OpenAI holiday launches always have that Christmas panic energy), and then… boom: GPT‑5.2 actually drops while we’re live.What makes this release feel significant isn’t “one benchmark went up.” It’s that OpenAI is clearly optimizing for the things that have become the frontier in 2025: long-horizon reasoning, agentic coding loops, long context reliability, and lower hallucination rates when browsing/tooling is involved.5.2 Instant, Thinking and Pro in ChatGPT and in the APIOpenAI shipped multiple variants, and even within those there are “levels” (medium/high/extra-high) that effectively change how much compute the model is allowed to burn. At the extreme end, you’re basically running parallel thoughts and selecting winners. That’s powerful, but also… very expensive.It’s very clearly aimed at the agentic world: coding agents that run in loops, tool-using research agents, and “do the whole task end-to-end” workflows where spending extra tokens is still cheaper than spending an engineer day.Benchmarks I’m not going to pretend benchmarks tell the full story (they never do), but the shape of improvements matters. GPT‑5.2 shows huge strength on reasoning + structured work.It hits 90.5% on ARC‑AGI‑1 in the Pro X‑High configuration, and 54%+ on ARC‑AGI‑2 depending on the setting. For context, ARC‑AGI‑2 is the one where everyone learns humility again.On math/science, this thing is flexing. We saw 100% on AIME 2025, and strong performance on FrontierMath tiers (with the usual “Tier 4 is where dreams go to die” vibe still intact). GPQA Diamond is up in the 90s too, which is basically “PhD trivia mode.”But honestly the most practically interesting one for me is GDPval (knowledge-work tasks: slides, spreadsheets, planning, analysis). GPT‑5.2 lands around 70%, which is a massive jump vs earlier generations. This is the category that translates directly into “is this model useful at my job.” - This is a bench that OpenAI launched only in September and back then, Opus 4.1 was a “measly” 47%! Talk about acceleration! Long context: MRCR is the sleeper highlightOn MRCR (multi-needle long-context retrieval), GPT‑5.2 holds up absurdly well even into 128k and beyond. The graph OpenAI shared shows GPT‑5.1 falling off a cliff as context grows, while GPT‑5.2 stays high much deeper into long contexts.If you’ve ever built a real system (RAG, agent memory, doc analysis) you know this pain: long context is easy to offer, hard to use well. If GPT‑5.2 actually delivers this in production, it’s a meaningful shift.Hallucinations: down (especially with browsing)One thing we called out on the show is that a bunch of user complaints in 2025 have basically collapsed into one phrase: “it hallucinates.” Even people who don’t know what a benchmark is can feel when a model confidently lies.OpenAI’s system card shows lower rates of major incorrect claims compared to GPT‑5.1, and lower “incorrect claims” overall when browsing is enabled. That’s exactly the direction they needed.Real-world vibes:We did the traditional “vibe tests” mid-show: generate a flashy landing page, do a weird engineering prompt, try some coding inside Cursor/Codex.Early testers broadly agree on the shape of the improvement. GPT‑5.2 is much stronger in reasoning, math, long‑context tasks, visual understanding, and multimodal workflows, with multiple reports of it successfully thinking for one to three hours on hard problems. Enterprise users like Box report faster execution and higher accuracy on real knowledge‑worker tasks, while researchers note that GPT‑5.2 Pro consistently outperforms the standard “Thinking” variant. The tradeoffs are also clear: creative writing still slightly favors Claude Opus, and the highest reasoning tiers can be slow and expensive. But as a general‑purpose reasoning model, GPT‑5.2 is now the strongest publicly available option.AI in space: Starcloud trains an LLM on an H100 in orbitThis story is peak 2025.Starcloud put an NVIDIA H100 on a satellite, trained Andrej Karpathy’s nanoGPT on Shakespeare, and ran inference on Gemma. There’s a viral screenshot vibe here that’s impossible to ignore: SSH into an H100… in space… with a US flag in the corner. It’s engineered excitement, and I’m absolutely here for it.But we actually had a real debate on the show: is “GPUs in space” just sci‑fi marketing, or does it make economic sense?Nisten made a compelling argument that power is the real bottleneck, not compute, and that big satellites already operate in the ~20kW range. If you can generate that power reliably with solar in orbit, the economics start looking less insane than you’d think. LDJ added the long-term land/power convergence argument: Earth land and grid power get scarcer/more regulated, while launch costs trend down—eventually the curves may cross.I played “voice of realism” for a minute: what happens when GPUs fail? It’s hard enough to swap a GPU in a datacenter, now imagine doing it in orbit. Cooling and heat dissipation become a different engineering problem too (radiators instead of fans). Networking is nontrivial. But also: we are clearly entering the era where people will try weird infra ideas because AI demand is pulling the whole economy.Big Company: MCP gets donated, OpenRouter drops a report on AIAgentic AI Foundation Lands at the Linux FoundationThis one made me genuinely happy.Block, Anthropic, and OpenAI came together to launch the Agentic AI Foundation under the Linux Foundation, donating key projects like MCP, AGENTS.md, and goose. This is exactly how standards should happen: vendor‑neutral, boring governance, lots of stakeholders.It’s not flashy work, but it’s the kind of thing that actually lets ecosystems grow without fragmenting. BTW, I was recording my podcast while Latent.Space were recording theirs in the same office, and they have a banger episode upcoming about this very topic! All I’ll say is Alessio Fanelli introduced me to David Soria Parra from MCP 👀 Watch out for that episode on Latent space dropping soon! OpenRouter’s “State of AI”: 100 Trillion Tokens of RealityOpenRouter and a16z dropped a massive report analyzing over 100 trillion tokens of real‑world usage. A few things stood out:Reasoning tokens now dominate. Above 50%, around 60% of all tokens since early 2025 are reasoning tokens. Remember when we went from “LLMs can’t do math” to reasoning models? That happened in about a year.Programming exploded. From 11% of usage early 2025 to over 50% recently. Claude holds 60% of the coding market. (at least.. on Open Router)Open source hit 30% market share, led by Chinese labs: DeepSeek (14T tokens), Qwen (5.59T), Meta LLaMA (3.96T).Context lengths grew massively. Average prompt length went from 1.5k to 6k+ tokens (4x growth), completions from 133 to 400 tokens (3x).The “Glass Slipper” effect. When users find a model that fits their use case, they stay loyal. Foundational early-user cohorts retain around 40% at month 5. Claude 4 Sonnet still had 50% retention after three months.Geography shift. Asia doubled to 31% of usage (China key), while North America is at 47%.Yam made a good point that we should be careful interpreting these graphs—they’re biased toward people trying new models, not necessarily steady usage. But the trends are clear: agentic, reasoning, and coding are the dominant use cases.Open Source Is Not Slowing Down (If Anything, It’s Accelerating)One of the strongest themes this week was just how fast open source is closing the gap — and in some areas, outright leading. We’re not talking about toy demos anymore. We’re talking about serious models, trained from scratch, hitting benchmarks that were frontier‑only not that long ago.Essential AI’s Rnj‑1: A Real Frontier 8B ModelThis one deserves real attention. Essential AI — led by Ashish Vaswani, yes Ashish from the original Transformers paper — released Rnj‑1, a pair of 8B open‑weight models trained fully from scratch. No distillation. No “just a fine‑tune.” This is a proper pretrain.What stood out to me isn’t just the benchmarks (though those are wild), but the philosophy. Rnj‑1 is intentionally focused on pretraining quality: data curation, code execution simulation, STEM reasoning, and agentic behaviors emerging during pretraining instead of being bolted on later with massive RL pipelines.In practice, that shows up in places like SWE‑bench Verified, where Rnj‑1 lands in the same ballpark as much larger closed models, and in math and STEM tasks where it punches way above its size. And remember: this is an 8B model you can actually run locally, quantize aggressively, and deploy without legal gymnastics thanks to its Apache 2.0 license.Mistral Devstral 2 + Vibe: Open Coding Goes HardMistral followed up last week’s momentum with Devstral 2, and Mistral Vibe! The headline numbers are: the 123B Devstral 2 model lands right at the top of open‑weight coding benchmarks, nearly matching Claude 3.5 Sonnet on SWE‑bench Verified. But what really excited the panel was the 24B Devstral Small 2, which hits high‑60s SWE‑bench scores while being runnable on consumer hardware.This is the kind of model you can realistically run

Hey yall, Alex here 🫡 Welcome to the first ThursdAI of December! Snow is falling in Colorado, and AI releases are falling even harder. This week was genuinely one of those “drink from the firehose” weeks where every time I refreshed my timeline, another massive release had dropped.We kicked off the show asking our co-hosts for their top AI pick of the week, and the answers were all over the map: Wolfram was excited about Mistral’s return to Apache 2.0, Yam couldn’t stop talking about Claude Opus 4.5 after a full week of using it, and Nisten came out of left field with an AWQ quantization of Prime Intellect’s model that apparently runs incredibly fast on a single GPU. As for me? I’m torn between Opus 4.5 (which literally fixed bugs that Gemini 3 created in my code) and DeepSeek’s gold-medal winning reasoning model.Speaking of which, let’s dive into what happened this week, starting with the open source stuff that’s been absolutely cooking. Open Source LLMsDeepSeek V3.2: The Whale Returns with Gold MedalsThe whale is back, folks! DeepSeek released two major updates this week: V3.2 and V3.2-Speciale. And these aren’t incremental improvements—we’re talking about an open reasoning-first model that’s rivaling GPT-5 and Gemini 3 Pro with actual gold medal Olympiad wins.Here’s what makes this release absolutely wild: DeepSeek V3.2-Speciale is achieving 96% on AIME versus 94% for GPT-5 High. It’s getting gold medals on IMO (35/42), CMO, ICPC (10/12), and IOI (492/600). This is a 685 billion parameter MOE model with MIT license, and it literally broke the benchmark graph on HMMT 2025—the score was so high it went outside the chart boundaries. That’s how you DeepSeek, basically.But it’s not just about reasoning. The regular V3.2 (not Speciale) is absolutely crushing it on agentic benchmarks: 73.1% on SWE-Bench Verified, first open model over 35% on Tool Decathlon, and 80.3% on τ²-bench. It’s now the second most intelligent open weights model and ranks ahead of Grok 4 and Claude Sonnet 4.5 on Artificial Analysis.The price is what really makes this insane: 28 cents per million tokens on OpenRouter. That’s absolutely ridiculous for this level of performance. They’ve also introduced DeepSeek Sparse Attention (DSA) which gives you 2-3x cheaper 128K inference without performance loss. LDJ pointed out on the show that he appreciates how transparent they’re being about not quite matching Gemini 3’s efficiency on reasoning tokens, but it’s open source and incredibly cheap.One thing to note: V3.2-Speciale doesn’t support tool calling. As Wolfram pointed out from the model card, it’s “designed exclusively for deep reasoning tasks.” So if you need agentic capabilities, stick with the regular V3.2.Check out the full release on Hugging Face or read the announcement.Mistral 3: Europe’s Favorite AI Lab Returns to Apache 2.0Mistral is back, and they’re back with fully open Apache 2.0 licenses across the board! This is huge news for the open source community. They released two major things this week: Mistral Large 3 and the Ministral 3 family of small models.Mistral Large 3 is a 675 billion parameter MOE with 41 billion active parameters and a quarter million (256K) context window, trained on 3,000 H200 GPUs. There’s been some debate about this model’s performance, and I want to address the elephant in the room: some folks saw a screenshot showing Mistral Large 3 very far down on Artificial Analysis and started dunking on it. But here’s the key context that Merve from Hugging Face pointed out—this is the only non-reasoning model on that chart besides GPT 5.1. When you compare it to other instruction-tuned (non-reasoning) models, it’s actually performing quite well, sitting at #6 among open models on LMSys Arena.Nisten checked LM Arena and confirmed that on coding specifically, Mistral Large 3 is scoring as one of the best open source coding models available. Yam made an important point that we should compare Mistral to other open source players like Qwen and DeepSeek rather than to closed models—and in that context, this is a solid release.But the real stars of this release are the Ministral 3 small models: 3B, 8B, and 14B, all with vision capabilities. These are edge-optimized, multimodal, and the 3B actually runs completely in the browser with WebGPU using transformers.js. The 14B reasoning variant achieves 85% on AIME 2025, which is state-of-the-art for its size class. Wolfram confirmed that the multilingual performance is excellent, particularly for German.There’s been some discussion about whether Mistral Large 3 is a DeepSeek finetune given the architectural similarities, but Mistral claims these are fully trained models. As Nisten noted, even if they used similar architecture (which is Apache 2.0 licensed), there’s nothing wrong with that—it’s an excellent architecture that works. Lucas Atkins later confirmed on the show that “Mistral Large looks fantastic... it is DeepSeek through and through architecture wise. But Kimi also does that—DeepSeek is the GOAT. Training MOEs is not as easy as just import deepseak and train.”Check out Mistral Large 3 and Ministral 3 on Hugging Face.Arcee Trinity: US-Trained MOEs Are BackWe had Lucas Atkins, CTO of Arcee AI, join us on the show to talk about their new Trinity family of models, and this conversation was packed with insights about what it takes to train MOEs from scratch in the US.Trinity is a family of open-weight MOEs fully trained end-to-end on American infrastructure with 10 trillion curated tokens from Datology.ai. They released Trinity-Mini (26B total, 3B active) and Trinity-Nano-Preview (6B total, 1B active), with Trinity-Large (420B parameters, 13B active) coming in mid-January 2026.The benchmarks are impressive: Trinity-Mini hits 84.95% on MMLU (0-shot), 92.1% on Math-500, and 65% on GPQA Diamond. But what really caught my attention was the inference speed—Nano generates at 143 tokens per second on llama.cpp, and Mini hits 157 t/s on consumer GPUs. They’ve even demonstrated it running on an iPhone via MLX Swift.I asked Lucas why it matters where models come from, and his answer was nuanced: for individual developers, it doesn’t really matter—use the best model for your task. But for Fortune 500 companies, compliance and legal teams are getting increasingly particular about where models were trained and hosted. This is slowing down enterprise AI adoption, and Trinity aims to solve that.Lucas shared a fascinating insight about why they decided to do full pretraining instead of just post-training on other people’s checkpoints: “We at Arcee were relying on other companies releasing capable open weight models... I didn’t like the idea of the foundation of our business being reliant on another company releasing models.” He also dropped some alpha about Trinity-Large: they’re going with 13B active parameters instead of 32B because going sparser actually gave them much faster throughput on Blackwell GPUs.The conversation about MOEs being cheaper for RL was particularly interesting. Lucas explained that because MOEs are so inference-efficient, you can do way more rollouts during reinforcement learning, which means more RL benefit per compute dollar. This is likely why we’re seeing labs like MiniMax go from their original 456B/45B-active model to a leaner 220B/10B-active model—they can get more gains in post-training by being able to do more steps.Check out Trinity-Mini and Trinity-Nano-Preview on Hugging Face, or read The Trinity Manifesto.OpenAI Code Red: Panic at the Disco (and Garlic?)It was ChatGPT’s 3rd birthday this week (Nov 30th), but the party vibes seem… stressful. Reports came out that Sam Altman has declared a “Code Red” at OpenAI.Why? Gemini 3.The user numbers don’t lie. ChatGPT apparently saw a 6% drop in daily active users following the Gemini 3 launch. Google’s integration is just too good, and their free tier is compelling.In response, OpenAI has supposedly paused “side projects” (ads, shopping bots) to focus purely on model intelligence and speed. Rumors point to a secret model codenamed “Garlic”—a leaner, more efficient model that beats Gemini 3 and Claude Opus 4.5 on coding reasoning, targeting a release in early 2026 (or maybe sooner if they want to save Christmas).Wolfram and Yam nailed the sentiment here: Integration wins. Wolfram’s family uses Gemini because it’s right there on the Pixel, controlling the lights and calendar. OpenAI needs to catch up not just on IQ, but on being helpful in the moment.Post the live show, OpenAI also finally added GPT 5.1 Codex Max we covered 2 weeks ago to their API and it’s now available in Cursor, for free, until Dec 11! Amazon Nova 2: Enterprise Push with Serious Agentic ChopsAmazon came back swinging with Nova 2, and the jump on Artificial Analysis is genuinely impressive—from around 30% to 61% on their index. That’s a massive improvement.The family includes Nova 2 Lite (7x cheaper, 5x faster than Nova Premier), Nova 2 Pro (93% on τ²-Bench Telecom, 70% on SWE-Bench Verified), Nova 2 Sonic (speech-to-speech with 1.39s time-to-first-audio), and Nova 2 Omni (unified text/image/video/speech with 1M token context window—you can upload 90 minutes of video!).Gemini 3 Deep Think ModeGoogle launched Gemini 3 Deep Think mode exclusively for AI Ultra subscribers, and it’s hitting some wild benchmarks: 45.1% on ARC-AGI-2 (a 2x SOTA leap using code execution), 41% on Humanity’s Last Exam, and 93.8% on GPQA Diamond. This builds on their Gemini 2.5 variants that earned gold medals at IMO and ICPC World Finals. The parallel reasoning approach explores multiple hypotheses simultaneously, but it’s compute-heavy—limited to 10 prompts per day at $77 per ARC-AGI-2 task.This Week’s Buzz: Mid-Training Evals are Here!A huge update from us at Weights & Biases this week: We launched LLM Evaluation Jobs. (Docs)If you are training models or finetuning, you usually wait until the end to run your expensive benchmarks. Now, directly inside W&B, you can trigger

Hey, Alex here, I recorded these conversations just in front of the AI Engineer auditorium, back to back, after these great folks gave their talks, and at the epitome of the most epic AI week we’ve seen since I started recording ThursdAI.This is less our traditional live recording, and more a real podcast-y conversation with great folks, inspired by Latent.Space. I hope you enjoy this format as much as I’ve enjoyed recording and editing it. AntiGravity with KevinKevin Hou and team just launched Antigravity, Google’s brand new Agentic IDE based on VSCode, and Kevin (second timer on ThursdAI) was awesome enough to hop on and talk about some of the product decisions they made, what makes Antigravity special and highlighted Artifacts as a completely new primitive. Gemini 3 in AI StudioIf you aren’t using Google’s AI Studio (ai.dev) then you’re missing out! We talk about AI Studio all the time on the show, and I’m a daily user! I generate most of my images with Nano Banana Pro in there, most of my Gemini conversations are happening there as well! Ammaar and Kat were so fun to talk to, as they covered the newly shipped “build mode” which allows you to vibe code full apps and experiences inside AI Studio, and we also covered Gemini 3’s features, multimodality understanding, UI capabilities. These folks gave a LOT of Gemini 3 demo’s so they know everything there is to know about this model’s capabilities! Tried new things with this one, multi camera angels, conversation with great folks, if you found this content valuable, please subscribe :) Topics Covered:* Inside Google’s new “AntiGravity” IDE* How the “Agent Manager” changes coding workflows* Gemini 3’s new multimodal capabilities* The power of “Artifacts” and dynamic memory* Deep dive into AI Studio updates & Vibe Coding* Generating 4K assets with Nano Banana ProTimestamps for your viewing convenience. 00:00 - Introduction and Overview01:13 - Conversation with Kevin Hou: Anti-Gravity IDE01:58 - Gemini 3 and Nano Banana Pro Launch Insights03:06 - Innovations in Anti-Gravity IDE06:56 - Artifacts and Dynamic Memory09:48 - Agent Manager and Multimodal Capabilities11:32 - Chrome Integration and Future Prospects20:11 - Conversation with Ammar and Kat: AI Studio Team21:21 - Introduction to AI Studio21:51 - What is AI Studio?22:52 - Ease of Use and User Feedback24:06 - Live Demos and Launch Week26:00 - Design Innovations in AI Studio30:54 - Generative UIs and Vibe Coding33:53 - Nano Banana Pro and Image Generation39:45 - Voice Interaction and Future Roadmap44:41 - Conclusion and Final ThoughtsLooking forward to seeing you on Thursday 🫡 P.S - I’ve recorded one more conversation during AI Engineer, and will be posting that soon, same format, very interesting person, look out for that soon! This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit sub.thursdai.news/subscribe

Hey ya’ll, Happy Thanskgiving to everyone who celebrates and thank you for being a subscriber, I truly appreciate each and every one of you!Just wrapped up the third (1, 2) Thanksgiving special Episode of ThursdAI, can you believe November is almost over? We had another banger week in AI, with a full feast of AI released, Anthropic dropped the long awaited Opus 4.5, which quickly became the top coding LLM, DeepSeek resurfaced with a math model, BFL and Tongyi both tried to take on Nano Banana, and Microsoft dropped a 7B computer use model in Open Source + Intellect 3 from Prime Intellect! With so much news to cover, we also had an interview with Ido Sal & Liad Yosef (their second time on the show!) about MCP-Apps, the new standard they are spearheading together with Anthropic, OpenAI & more! Exciting episode, let’s get into it! (P.S - I started generating infographics, so the show became much more visual, LMK if you like them) ThursdAI - I put a lot of work on a weekly basis to bring you the live show, podcast and a sourced newsletter! Please subscribe if you find this content valuable!Anthropic’s Opus 4.5: The “Premier Intelligence” Returns (Blog)Folks, Anthropic absolutely cooked. After Sonnet and Haiku had their time in the sun, the big brother is finally back. Opus 4.5 launched this week, and it is reclaiming the throne for coding and complex agentic tasks.First off, the specs are monstrous. It hits 80.9% on SWE-bench Verified, topping GPT-5.1 (77.9%) and Gemini 3 Pro (76.2%). But the real kicker? The price! It is now $5 per million input tokens and $25 per million output—literally one-third the cost of the previous Opus.Yam, our resident coding wizard, put it best during the show: “Opus knows a lot of tiny details about the stack that you didn’t even know you wanted... It feels like it can go forever.” Unlike Sonnet, which sometimes spirals or loses context on extremely long tasks, Opus 4.5 maintains coherence deep into the conversation.Anthropic also introduced a new “Effort” parameter, allowing you to control how hard the model thinks (similar to o1 reasoning tokens). Set it to high, and you get massive performance gains; set it to medium, and you get Sonnet-level performance at a fraction of the token cost. Plus, they’ve added Tool Search (cutting enormous token overhead for agents with many tools) and Programmatic Tool Calling, which effectively lets Opus write and execute code loops to manage data.If you are doing heavy software engineering or complex automations, Opus 4.5 is the new daily driver.📱 The Agentic Web: MCP Apps & MCP-UI StandardSpeaking of MCP updates, Can you believe it’s been exactly one year since the Model Context Protocol (MCP) launched? We’ve been “MCP-pilled” for a while, but this week, the ecosystem took a massive leap forward.We brought back our friends Ido and Liad, the creators of MCP-UI, to discuss huge news: MCP-UI has been officially standardized as MCP Apps. This is a joint effort adopted by both Anthropic and OpenAI.Why does this matter? Until now, when an LLM used a tool (like Spotify or Zillow), the output was just text. It lost the brand identity and the user experience. With MCP Apps, agents can now render full, interactive HTML interfaces directly inside the chat! Ido and Liad explained that they worked hard to avoid an “iOS vs. Android” fragmentation war. Instead of every lab building their own proprietary app format, we now have a unified standard for the “Agentic Web.” This is how AI stops being a chatbot and starts being an operating system.Check out the standard at mcpui.dev.🦃 The Open Source Thanksgiving FeastWhile the big labs were busy, the open-source community decided to drop enough papers and weights to feed us for a month.Prime Intellect unveils INTELLECT-3, a 106B MoE (X, HF, Blog, Try It)Prime Intellect releases INTELLECT-3, a 106B parameter Mixture-of-Experts model (12B active params) based on GLM-4.5-Air, achieving state-of-the-art performance for its size—including ~90% on AIME 2024/2025 math contests, 69% on LiveCodeBench v6 coding, 74% on GPQA-Diamond reasoning, and 74% on MMLU-Pro—outpacing larger models like DeepSeek-R1. Trained over two months on 512 H200 GPUs using their fully open-sourced end-to-end stack (PRIME-RL async trainer, Verifiers & Environments Hub, Prime Sandboxes), it’s now hosted on Hugging Face, OpenRouter, Parasail, and Nebius, empowering any team to scale frontier RL without big-lab resources. Especially notable is their very detailed release blog, covering how a lab that previously trained 32B, finetunes a monster 106B MoE model! Tencent’s HunyuanOCR: Small but Mighty (X, HF, Github, Blog)Tencent released HunyuanOCR, a 1 billion parameter model that is absolutely crushing benchmarks. It scored 860 on OCRBench, beating massive models like Qwen3-VL-72B. It’s an end-to-end model, meaning no separate detection and recognition steps. Great for parsing PDFs, docs, and even video subtitles. It’s heavily restricted (no EU/UK usage), but technically impressive.Microsoft’s Fara-7B: On-Device Computer UseMicrosoft quietly dropped Fara-7B, a model fine-tuned from Qwen 2.5, specifically designed for computer use agentic tasks. It hits 73.5% on WebVoyager, beating OpenAI’s preview models, all while running locally on-device. This is the dream of a local agent that can browse the web for you, click buttons, and book flights without sending screenshots to the cloud.DeepSeek-Math-V2: open-weights IMO-gold math LLM (X, HF)DeepSeek-Math-V2 is a 685B-parameter, Apache-2.0 licensed, open-weights mathematical reasoning model claiming gold-medal performance on IMO 2025 and CMO 2024, plus a near-perfect 118/120 on Putnam 2024. Nisten did note some limitations—specifically that the context window can get choked up on extremely long, complex proofs—but having an open-weight model of this caliber is a gift to researchers everywhere.🐝 This Week’s Buzz: Serverless LoRA InferenceA huge update from us at Weights & Biases! We know fine-tuning is powerful, but serving those fine-tunes can be a pain and expensive. We just launched Serverless LoRA Inference.This means you can upload your small LoRA adapters (which you can train cheaply) to W&B Artifacts, and we will serve them instantly on CoreWeave GPUs on top of a base model. No cold starts, no dedicated expensive massive GPU instances for just one adapter.I showed a demo of a “Mocking SpongeBob” model I trained in 25 minutes. It just adds that SaRcAsTiC tExT style to the Qwen 2.5 base model. You pass the adapter ID in the API call, and boom—customized intelligence instantly. You can get more details HERE and get started with your own LORA in this nice notebook the team made! 🎨 Visuals: Image & Video Generation ExplosionFlux.2: The Multi-Reference Image Creator from BFL (X, HF, Blog)Black Forest Labs released Flux.2, a series of models including a 32B Flux 2[DEV]. The killer feature here is Multi-Reference Editing. You can feed it up to 10 reference images to maintain character consistency, style, or specific objects. It also outputs native 4-megapixel images.Honestly, the launch timing was rough, coming right after Google’s Nano Banana Pro and alongside Z-Image, but for precise, high-res editing, this is a serious tool.Tongyi drops Z-Image Turbo: 6B single-stream DiT lands sub‑second, 8‑step text‑to‑image (GitHub, Hugging Face)Alibaba’s Tongyi Lab released Z-Image Turbo, a 6B parameter model that generates images in sub-second time on H800s (and super fast on consumer cards).I built a demo to show just how fast this is. You know that “Infinite Craft“ game? I hooked it up to Z-Image Turbo so that every time you combine elements (like Pirate + Ghost), it instantly generates the image for “Ghost Pirate.” It changes the game completely when generation is this cheap and fast.HunyuanVideo 1.5 – open video gets very realTencent also shipped HunyuanVideo 1.5, which they market as “the strongest open‑source video generation model.” For once, the tagline isn’t entirely hype.Under the hood it’s an 8.3B‑parameter Diffusion Transformer (DiT) model with a 3D causal VAE in front. The VAE compresses videos aggressively in both space and time, and the DiT backbone models that latent sequence.The important bits for you and me:* It generates 5–10 second clips at 480p/720p with good motion coherence and physics.* With FP16 or FP8 configs you can run it on a single consumer GPU with around 14GB VRAM.* There’s a built‑in path to upsample to 1080p for “real” quality.LTX Studio Retake: Photoshop for Video (X, Try It)For the video creators, LTX Studio launched Retake. This isn’t just “regenerate video.” This allows you to select a specific 2-second segment of a video, change the dialogue (keeping the voice!), change the emotion, or edit the action, all for like $0.10. It blends it perfectly back into the original clip. We are effectively getting a “Director Mode” for AI video where you can fix mistakes without rolling the dice on a whole new generation.A secret new model on the Arena called Whisper Thunder beats even Veo 3?This was a surprise of the week, while new video models get released often, Veo 3 has been the top one for a while, and now we’re getting a reshuffling of the video giants! But... we don’t yet know who this video model is from! You can sometimes get its generations at the Artificial Analysis video arena here, and the outputs look quite awesome! Thanksgiving reflections from the ThursdAI teamAs we wrapped up the show, Wolfram suggested we take a moment to think about what we’re thankful for in AI, and I think that’s a perfect note to end on.Wolfram put it well: he’s thankful for everyone contributing to this wonderful community—the people releasing models, creating open source tools, writing tutorials, sharing knowledge. It’s not just about the money; it’s about the love of learning and building together.Yam highlighted something I think is crucial: we’ve reached a point where there’s no real competition between o

Hey everyone, Alex here 👋I’m writing this one from a noisy hallway at the AI Engineer conference in New York, still riding the high (and the sleep deprivation) from what might be the craziest week we’ve ever had in AI.In the span of a few days:Google dropped Gemini 3 Pro, a new Deep Think mode, generative UIs, and a free agent-first IDE called Antigravity.xAI shipped Grok 4.1, then followed it up with Grok 4.1 Fast plus an Agent Tools API.OpenAI answered with GPT‑5.1‑Codex‑Max, a long‑horizon coding monster that can work for more than a day, and quietly upgraded ChatGPT Pro to GPT‑5.1 Pro.Meta looked at all of that and said “cool, we’ll just segment literally everything and turn photos into 3D objects” with SAM 3 and SAM 3D.Robotics folks dropped a home robot trained with almost no robot data.And Google, just to flex, capped Thursday with Nano Banana Pro, a 4K image model and a provenance system while we were already live! For the first time in a while it doesn’t just feel like “new models came out.” It feels like the future actually clicked forward a notch.This is why ThursdAI exists. Weeks like this are basically impossible to follow if you have a day job, so my co‑hosts and I do the no‑sleep version so you don’t have to. Plus, being at AI Engineer makes it easy to get super high quality guests so this week we had 3 folks join us, Swyx from Cognition/Latent Space, Thor from DeepMind (on his 3rd day) and Dominik from OpenAI! Alright, deep breath. Let’s untangle the week.TL;DR If you only skim one section, make it this one (links in the end):* Google* Gemini 3 Pro: 1M‑token multimodal model, huge reasoning gains - new LLM king* ARC‑AGI‑2: 31.11% (Pro), 45.14% (Deep Think) – enormous jumps* Antigravity IDE: free, Gemini‑powered VS Code fork with agents, plans, walkthroughs, and browser control* Nano Banana Pro: 4K image generation with perfect text + SynthID provenance; dynamic “generative UIs” in Gemini* xAI* Grok 4.1: big post‑training upgrade – #1 on human‑preference leaderboards, much better EQ & creative writing, fewer hallucinations* Grok 4.1 Fast + Agent Tools API: 2M context, SOTA tool‑calling & agent benchmarks (Berkeley FC, T²‑Bench, research evals), aggressive pricing and tight X + web integration* OpenAI* GPT‑5.1‑Codex‑Max: “frontier agentic coding” model built for 24h+ software tasks with native compaction for million‑token sessions; big gains on SWE‑Bench, SWE‑Lancer, TerminalBench 2* GPT‑5.1 Pro: new “research‑grade” ChatGPT mode that will happily think for minutes on a single query* Meta* SAM 3: open‑vocabulary segmentation + tracking across images and video (with text & exemplar prompts)* SAM 3D: single‑image → 3D objects & human bodies; surprisingly high‑quality 3D from one photo* Robotics* Sunday Robotics – ACT‑1 & Memo: home robot foundation model trained from a $200 skill glove instead of $20K teleop rigs; long‑horizon household tasks with solid zero‑shot generalization* Developer Tools* Antigravity and Marimo’s VS Code / Cursor extension both push toward agentic, reactive dev workflowsLive from AI Engineer New York: Coding Agents Take Center StageWe recorded this week’s show on location at the AI Engineer Summit in New York, inside a beautiful podcast studio the team set up right on the expo floor. Huge shout out to Swyx, Ben, and the whole AI Engineer crew for that — last time I was balancing a mic on a hotel nightstand, this time I had broadcast‑grade audio while a robot dog tried to steal the show behind us.This year’s summit theme is very on‑the‑nose for this week: coding agents.Everywhere you look, there’s a company building an “agent lab” on top of foundation models. Amp, Cognition, Cursor, CodeRabbit, Jules, Google Labs, all the open‑source folks, and even the enterprise players like Capital One and Bloomberg are here, trying to figure out what it means to have real software engineers that are partly human and partly model.Swyx framed it nicely when he said that if you take “vertical AI” seriously enough, you eventually end up building an agent lab. Lawyers, healthcare, finance, developer tools — they all converge on “agents that can reason and code.”The big labs heard that theme loud and clear. Almost every major release this week is about agents, tools, and long‑horizon workflows, not just chat answers.Google Goes All In: Gemini 3 Pro, Antigravity, and the Agent RevolutionLet’s start with Google because, after years of everyone asking “where’s Google?” in the AI race, they showed up this week with multiple bombshells that had even the skeptics impressed.Gemini 3 Pro: Multimodal Intelligence That Actually DeliversGoogle finally released Gemini 3 Pro, and the numbers are genuinely impressive. We’re talking about a 1 million token context window, massive benchmark improvements, and a model that’s finally competing at the very top of the intelligence charts. Thor from DeepMind joined us on the show (literally on day 3 of his new job!) and you could feel the excitement.The headline numbers: Gemini 3 Pro with Deep Think mode achieved 45.14% on ARC-AGI-2—that’s roughly double the previous state-of-the-art on some splits. For context, ARC-AGI has been one of those benchmarks that really tests genuine reasoning and abstraction, not just memorization. The standard Gemini 3 Pro hits 31.11% on the same benchmark, both scores are absolutely out of this world in Arc! On GPQA Diamond, Gemini 3 Pro jumped about 10 points compared to prior models. We’re seeing roughly 81% on MMLU-Pro, and the coding performance is where things get really interesting—Gemini 3 Pro is scoring around 56% on SciCode, representing significant improvements in actual software engineering tasks.But here’s what made Ryan from Amp switch their default model to Gemini 3 Pro immediately: the real-world usability. Ryan told us on the show that they’d never switched default models before, not even when GPT-5 came out, but Gemini 3 Pro was so noticeably better that they made it the default on Tuesday. Of course, they hit rate limits almost immediately (Google had to scale up fast!), but those have since been resolved.Antigravity: Google’s Agent-First IDEThen Google dropped Antigravity, and honestly, this might be the most interesting part of the whole release. It’s a free IDE (yes, free!) that’s basically a fork of VS Code, but reimagined around agents rather than human-first coding.The key innovation here is something they call the “Agent Manager”—think of it like an inbox for your coding agents. Instead of thinking in folders and files, you’re managing conversations with agents that can run in parallel, handle long-running tasks, and report back when they need your input.I got early access and spent time playing with it, and here’s what blew my mind: you can have multiple agents working on different parts of your codebase simultaneously. One agent fixing bugs, another researching documentation, a third refactoring your CSS—all at once, all coordinated through this manager interface.The browser integration is crazy too. Antigravity can control Chrome directly, take screenshots and videos of your app, and then use those visuals to debug and iterate. It’s using Gemini 3 Pro for the heavy coding, and even Nano Banana for generating images and assets. The whole thing feels like it’s from a couple years in the future.Wolfram on the show called out how good Gemini 3 is for creative writing too—it’s now his main model, replacing GPT-4.5 for German language tasks. The model just “gets” the intention behind your prompts rather than following them literally, which makes for much more natural interactions.Nano Banana Pro: 4K Image Generation With ThinkingAnd because Google apparently wasn’t done announcing things, they also dropped Nano Banana Pro on Thursday morning—literally breaking news during our live show. This is their image generation model that now supports 4K resolution and includes “thinking” traces before generating.I tested it live by having it generate an infographic about all the week’s AI news (which you can see on the top), and the results were wild. Perfect text across the entire image (no garbled letters!), proper logos for all the major labs, and compositional understanding that felt way more sophisticated than typical image models. The file it generated was 8 megabytes—an actual 4K image with stunning detail.What’s particularly clever is that Nano Banana Pro is really Gemini 3 Pro doing the thinking and planning, then handing off to Nano Banana for the actual image generation. So you get multimodal reasoning about your request, then production-quality output. You can even upload reference images—up to 14 of them—and it’ll blend elements while maintaining consistency.Oh, and every image is watermarked with SynthID (Google’s invisible watermarking tech) and includes C2PA metadata, so you can verify provenance. This matters as AI-generated content becomes more prevalent.Generative UIs: The Future of InterfacesOne more thing Google showed off: generative UIs in the Gemini app. Wolfram demoed this for us, and it’s genuinely impressive. Instead of just text responses, Gemini can generate full interactive mini-apps on the fly—complete dashboards, data visualizations, interactive widgets—all vibe-coded in real time.He asked for “four panels of the top AI news from last week” and Gemini built an entire news dashboard with tabs, live market data (including accurate pre-market NVIDIA stats!), model comparisons, and clickable sections. It pulled real information, verified facts, and presented everything in a polished UI that you could interact with immediately.This isn’t just a demo—it’s rolling out in Gemini now. The implication is huge: we’re moving from static responses to dynamic, contextual interfaces generated just-in-time for your specific need.xAI Strikes Back: Grok 4.1 and the Agent Tools APINot to be outdone, xAI released Grok 4.1 at the start of the week, briefly claimed the #1 spot on LMArena (at 1483 Elo, not 2nd to

Hey, this is Alex! We’re finally so back! Tons of open source releases, OpenAI updates GPT and a few breakthroughs in audio as well, makes this a very dense week! Today on the show, we covered the newly released GPT 5.1 update, a few open source releases like Terminal Bench and Project AELLA (renamed OASSAS), and Baidu’s Ernie 4.5 VL that shows impressive visual understanding! Also, chatted with Paul from 11Labs and Dima Duev from the wandb SDK team, who brought us a delicious demo of LEET, our new TUI for wandb! Tons of news coverage, let’s dive in 👇 (as always links and show notes in the end) Open Source AILet’s jump directly into Open Source as this week has seen some impressive big company models. Terminal-Bench 2.0 - a harder, highly‑verified coding and terminal benchmark (X, Blog, Leaderboard)We opened with Terminal‑Bench 2.0 plus its new harness, Harbor, because this is the kind of benchmark we’ve all been asking for. Terminal‑Bench focuses on agentic coding in a real shell. Version 2.0 is a hard set of 89 terminal tasks, each one painstakingly vetted by humans and LLMs to make sure it’s solvable and realistic. Think “I checked out master and broke my personal site, help untangle the git mess” or “implement GPT‑2 code golf with the fewest characters.” On the new leaderboard, top agents like Warp’s agentic console and Codex CLI + GPT‑5 sit around fifty percent success. That number is exactly what excites me: we’re nowhere near saturation. When everyone is in the 90‑something range, tiny 0.1 improvements are basically noise. When the best models are at fifty percent, a five‑point jump really means something.A huge part of our conversation focused on reproducibility. We’ve seen other benchmarks like OSWorld turn out to be unreliable, with different task sets and non‑reproducible results making scores incomparable. Terminal‑Bench addresses this with Harbor, a harness designed to run sandboxed, containerized agent rollouts at scale in a consistent environment. This means results are actually comparable. It’s a ton of work to build an entire evaluation ecosystem like this, and with over a thousand contributors on their Discord, it’s a fantastic example of a healthy, community‑driven effort. This is one to watch! Baidu’s ERNIE‑4.5‑VL “Thinking”: a 3B visual reasoner that punches way up (X, HF, GitHub)Next up, Baidu dropped a really interesting model, ERNIE‑4.5‑VL‑28B‑A3B‑Thinking. This is a compact, 3B active‑parameter multimodal reasoning model focused on vision, and it’s much better than you’d expect for its size. Baidu’s own charts show it competing with much larger closed models like Gemini‑2.5‑Pro and GPT‑5‑High on a bunch of visual benchmarks like ChartQA and DocVQA.During the show, I dropped a fairly complex chart into the demo, and ERNIE‑4.5‑VL gave me a clean textual summary almost instantly—it read the chart more cleanly than I could. The model is built to “think with images,” using dynamic zooming and spatial grounding to analyze fine details. It’s released under an Apache‑2.0 license, making it a serious candidate for edge devices, education, and any product where you need a cheap but powerful visual brain.Open Source Quick Hits: OSSAS, VibeThinker, and Holo TwoWe also covered a few other key open-source releases. Project AELLA was quickly rebranded to OSSAS (Open Source Summaries At Scale), an initiative to make scientific literature machine‑readable. They’ve released 100k paper summaries, two fine-tuned models for the task, and a 3D visualizer. It’s a niche but powerful tool if you’re working with massive amounts of research. (X, HF)WeiboAI (from the Chinese social media company) released VibeThinker‑1.5B, a tiny 1.5B‑parameter reasoning model that is making bold claims about beating the 671B DeepSeek R1 on math benchmarks. We discussed the high probability of benchmark contamination, especially on tests like AIME24, but even with that caveat, getting strong chain‑of‑thought math out of a 1.5B model is impressive and useful for resource‑constrained applications. (X, HF, Arxiv)Finally, we had some breaking news mid‑show: H Company released Holo Two, their next‑gen multimodal agent for controlling desktops, websites, and mobile apps. It’s a fine‑tune of Qwen3‑VL and comes in 4B and 8B Apache‑2.0 licensed versions, pushing the open agent ecosystem forward. (X, Blog, HF)ThursdAI - Recaps of the most high signal AI weekly spaces is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.Big Companies & APIsGPT‑5.1: Instant vs Thinking, and a new personality barThe biggest headline of the week was OpenAI shipping GPT‑5.1, and this was a hot topic of debate on the show. The update introduces two modes: “Instant” for fast, low‑compute answers, and “Thinking” for deeper reasoning on hard problems. OpenAI claims Instant mode uses 57% fewer tokens on easy tasks, while Thinking mode dedicates 71% more compute to difficult ones. This adaptive approach is a smart evolution.The release also adds a personality dropdown with options like Professional, Friendly, Quirky, and Cynical, aiming for a more “warm” and customizable experience. Yam and I felt this was a step in the right direction, as GPT‑5 could often feel a bit cold and uncommunicative. However, Wolfram had a more disappointing experience, finding that GPT‑5.1 performed significantly worse on his German grammar and typography tasks compared to GPT‑4 or Claude Sonnet 4.5. It’s a reminder that “upgrades” can be subjective and task‑dependent.Since the show was recorded, GPT 5.1 is also released in the API and they have published a prompting guide and some evals! With some significant jumps across SWE-bench verified and GPQA Diamond! We’ll be testing this model out all week. The highlight for this model is the creative writing, it was made public that this model was being tested on OpenRouter as Polaris-alpha and that one tops the eqbench creative writing benchmarks beating Sonnet 4.5 and Gemini! Grok‑4 Fast: 2M context and a native X superpowerGrok‑4 Fast from xAI apparenly quietly got a substantial upgrade to a 2M‑token context window, but the most interesting part is its unique integration with X. The API version has access to internal tools for semantic search over tweets, retrieving top quote tweets, and understanding embedded images and videos. I’ve started using it as a research agent in my show prep, and it feels like having a research assistant living inside X’s backend—something you simply can’t replicate with public tools.I still have my gripes about their “stealth upgrade” versioning strategy, which makes rigorous evaluation difficult, but as a practical tool, Grok‑4 Fast is incredibly powerful. It’s also surprisingly fast and cost‑effective, holding its own against other top models on benchmarks while offering a superpower that no one else has.Google SIMA 2: Embodied Agents in Virtual WorldsGoogle’s big contribution this week was SIMA 2, DeepMind’s latest embodied agent for 3D virtual worlds. SIMA lives inside real games like No Man’s Sky and Goat Simulator, seeing the screen and controlling the game via keyboard and mouse, using Gemini as its reasoning brain. Demos showed it following complex, sketch‑based instructions, like finding an object that looks like a drawing of a spaceship and jumping on top of it.When you combine this with Genie 3—Google’s world model that can generate playable environments from a single image—you see the bigger picture: agents that learn physics, navigation, and common sense by playing in millions of synthetic worlds. We’re not there yet, but the pieces are clearly being assembled. We also touched on the latest Gemini Live voice upgrade, which users are reporting feels much more natural and responsiveMore Big Company News: Qwen Deep Research, Code Arena, and CursorWe also briefly covered Qwen’s new Deep Research feature, which offers an OpenAI‑style research agent inside their ecosystem. LMSYS launched Blog, a fantastic live evaluation platform where models build real web apps agentically, with humans voting on the results. And in the world of funding, the AI‑native code editor Cursor raised a staggering $2.3 billion, a clear sign that AI is becoming the default way developers interact with code.This Week’s Buzz: W&B LEET – a terminal UI that sparks joyFor this week’s buzz, I brought on Dima Duev from our SDK team at Weights & Biases to show off a side project that has everyone at the company excited: LEET, the Lightweight Experiment Exploration Tool. Imagine you’re training on an air‑gapped HPC cluster, living entirely in your terminal. How do you monitor your runs? With LEET.You run your training script in W&B offline mode, and in another terminal, you type wandb beta leet. Your terminal instantly turns into a full TUI dashboard with live metric plots, system stats, and run configs. You can zoom into spikes in your loss curve, filter metrics, and see everything updating in real time, all without a browser or internet connection. It’s one of those tools that just sparks joy. It ships with the latest wandb SDK (v0.23.0+), so just upgrade and give it a try! Voice & Audio: Scribe v2 Realtime and Omnilingual ASRElevenLabs Scribe v2 Realtime: ASR built for agents (X, Announcement, Demo)We’ve talked a lot on this show about ElevenLabs as “the place you go to make your AI talk.” This week, they came for the other half of the conversation. Paul Asjes from ElevenLabs joined us to walk through Scribe v2 Realtime, their new low‑latency speech‑to‑text model. If you’re building a voice agent, you need ears, a brain, and a mouth. ElevenLabs already nailed the mouth, and now they’ve built some seriously good ears.Scribe v2 Realtime is designed to run at around 150 milliseconds median latency, across more than ninety languages. Watching Paul’s live demo, it felt comfortably real‑time. When he switched from English to Dutch mid‑sentence, the system just followed along

Hey, Alex here! Quick note, while preparing for this week, I posted on X that I don’t remember such a quiet week in AI since I started doing ThursdAI regularly, but then 45 min before the show started, Kimi dropped a SOTA oss reasoning model, turning a quiet week into an absolute banger. Besides Kimi, we covered the updated MCP thinking from Anthropic, and had Kenton Varda from cloudflare as a guest to talk about Code Mode, chatted about Windsurf and Cursor latest updates and covered OpenAI’s insane deals. Also, because it was a quiet week, I figured I’d use the opportunity to create an AI powered automation, and used N8N for that, and shared it on the stream, so if you’re interested in automating with AI with relatively low code, this episode is for you. Let’s dive inThursdAI - Recaps of the most high signal AI weekly spaces is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.Kimi K2 Thinking is Here and It’s a 1 Trillion Parameter Beast! (X, HF, Tech Blog)Let’s start with the news that got everyone’s energy levels skyrocketing right as we went live. Moonshot AI dropped Kimi K2 Thinking, an open-source, 1 trillion-parameter Mixture-of-Experts (MoE) model, and it’s an absolute monster.This isn’t just a numbers game; Kimi K2 Thinking is designed from the ground up to be a powerful agent. With just around 32 billion active parameters during inference, a massive 256,000 token context window, and an insane tool-calling capacity. They’re claiming it can handle 200-300 sequential tool calls without any human intervention. The benchmarks are just as wild. On the Humanities Last Exam (HLE), they’re reporting a score of 44.9%, beating out both GPT-5 and Claude 4.5 Thinking. While it doesn’t quite top the charts on SWE-bench verified, it’s holding its own against the biggest closed-source models out there. Seeing an open-source model compete at this level is incredibly exciting.During the show, we saw some truly mind-blowing demos, from a beautiful interactive visualization of gradient descent to a simulation of a virus attacking cells, all generated by the model. The model’s reasoning traces, which are exposed through the API, also seem qualitatively different from other models, showing a deep and thoughtful process. My co-hosts and I were blown away. The weights and a very detailed technical report are available on Hugging Face, so you can dive in and see for yourself. Shout out to the entire Moonshot AI team for this incredible release!Other open source updates from this week* HuggingFace released an open source “Smol Training Playbook” on training LLMs, it’s a 200+ interactive beast with visualizations, deep dives into pretraining, dataset, postraining and more! (HF)* Ai2 launches OlmoEarth — foundation models + open, end-to-end platform for fast, high-resolution Earth intelligence (X, Blog)* LongCat-Flash-Omni — open-source omni-modal system with millisecond E2E spoken interaction, 128K context and a 560B ScMoE backbone (X, HF, Announcement)Big Tech’s Big Moves: Apple, Amazon, and OpenAIThe big companies were making waves this week, starting with a blockbuster deal that might finally make Siri smart. Apple is reportedly will be paying Google around $1 billion per year to license a custom 1.2 trillion-parameter version of Gemini to power a revamped Siri.This is a massive move. The Gemini model will run on Apple’s Private Cloud Compute, keeping user data walled off from Google, and will handle Siri’s complex summarizer and planner functions. After years of waiting for Apple to make a significant move in GenAI, it seems they’re outsourcing the heavy lifting for now while they work to catch up with their own in-house models. As a user, I don’t really care who builds the model, as long as Siri stops being dumb!In more dramatic news, Perplexity revealed that Amazon sent them a legal threat to block their Comet AI assistant from shopping on Amazon.com. This infuriated me. My browser is my browser, and I should be able to use whatever tools I want to interact with the web. Perplexity took a strong stand with their blog post, “Bullying is Not Innovation,” arguing that user agents are distinct from scrapers and act on behalf of the user with their own credentials. An AI assistant is just that—an assistant. It shouldn’t matter if I ask my wife or my AI to buy something for me on Amazon. This feels like a move by Amazon to protect its ad revenue at the expense of user choice and innovation, and I have to give major props to Perplexity for being so transparent and fighting back.Finally, OpenAI continues its quest for infinite compute, announcing a multi-year strategic partnership with AWS. This comes on top of massive deals with NVIDIA, Microsoft, Oracle, and others, bringing their total commitment to compute into the trillions of dollars. It’s getting to a point where OpenAI seems “too big to fail,” as any hiccup could have serious repercussions for the entire tech economy, which is now heavily propped up by AI investment. Sam has clarified that they don’t think OpenAI wants to be too big to fail in a recent post on X, and that the recent miscommunications around the US government backstopping OpenAI’s infrastructure bailouts were taken out of context. 🤔 Coding with AI: The Evolution of MCP and New Dev ToolsThis week, we kicked off a new segment on the show: Coding with AI! Essentially realizing that we talk about AI coding a LOT, and decided to add a dedicated corner to it! And we started with a fascinating development in the world of agentic tooling. Anthropic published a blog post arguing that the standard way of using the Model Context Protocol (MCP) — by loading full tool definitions into the context window — is inefficient.Their solution? Have LLMs write code to interact with tools instead. This approach can slash token usage by over 98% in some cases. This idea sounded familiar, and that’s because Cloudflare had already explored it with a feature called “Code Mode.” We were lucky enough to have Kenton Varda, one of the authors of the Code Mode post and head of engineering for Cloudflare Workers, join us to discuss this shift.Kenton explained that LLMs are trained on vast amounts of code, making it a more “native language” for them than the artificial construct of tool calls. By generating code, agents can chain multiple tool calls together, process intermediate results, and operate much more efficiently without sending everything back through the neural network. While MCP still provides crucial standardization for discovering and authorizing tools, this “code execution” pattern seems to be the way forward for building more powerful and scalable agents.Windsurfs CodeMaps and Cursor multi agent executionsIn other coding with AI news, Windsurf has pushed an incredible feature, called CodeMaps. They will use their SWE-1 model to (quickly) generate Codemaps that will expalins a code-base to you, in a visual way. What starts where and goes where. It’s really useful to understand a new codebase or re-understand one you forgot about already! You can even chat with codemaps, to see if your overall system’s design is solid! Great addition that I’m sure will help many folks adopt Windsurf! And Cursor, another popular AI-native IDE, released a super-performant in-IDE browser and a wild multi-agent feature that queries multiple LLMs in parallel and then synthesizes their answers.This Week’s TutorialI finally got around to building some serious automations for ThursdAI, and folks, N8N has been a game-changer. What used to take me 30+ minutes of manual work now happens automatically in the background.Here’s what I built: A Telegram bot that takes Twitter/X links, fetches the tweets and all linked content, uses AI agents to extract and summarize the information, and then posts it to our announcement channel and my notes app. The coolest part? I built this whole thing in about 4 hours with the help of Atlas browser and GPT-5 literally telling me what to do at each step.During the show, we even live-tested swapping out GPT-4o-mini for Kimi K2 - took literally 30 seconds to connect via OpenRouter. I went through my node and explains how this all works on the show, so if you’ve wanted to learn about n8n, check it out starting around 01:13:00. If you want to see how my automation turned out, it will be posting all my links to the new telegram channel t.me/thursdai_news (expect it to be messy at first as I’m testing out the automation) Robotics - Xpeng’s “Iron” humanoid: big vibes, few specsAnother week, another humanoid robot that is supposedly “coming” in 2026! A humanoid from Xpeng went viral this week, marketed as “the most human‑like” robot with soft skin, bionic muscles, customizable sexes (yes, really, they have a woman humanoid), something called a VLT brain, and a 2026 production goal. Here’s what we didn’t get: a spec sheet. No DOF, speed, payload, compute TOPS, battery capacity, runtime, or safety pathway. No pricing, manufacturing strategy, or clear target markets. In other words: lots of sizzle, no steak.Apparently, there was folks thinking Xpend pulled an Elon and put a human in a robot suit, making the CEO do the “we’ll cut a part of the soft skin to expose the robot underneath so you don’t think we’re lying” stunt. Which I agree, was very effective. But, If Xpeng is serious, the next thing we’ll see should be a crisp engineering document: joints, actuation, sensors, compute, and a locomotion/manipulation demo with independent measurements. Until then, treat this as a branding salvo and a reminder that the humanoid category is still sorting itself into “industrial payload first” versus “human likeness first” approaches. Voice & AudioMaya‑1: open‑source voice design from natural languageWe highlighted Maya‑1, a 3B Llama‑backboned TTS system designed to generate voices from natural language descriptions. Instead of picking from a menu, you describe the voice—

Hey, it’s Alex! Happy Halloween friends! I’m excited to bring you this weeks (spooky) AI updates! We started the show today with MiniMax M2, the currently top Open Source LLM, with an interview with their head of eng, Skyler Miao, continued to dive into OpenAIs completed restructuring into a non-profit and a PBC, including a deep dive into a live stream Sam Altman had, with a ton of spicy details, and finally chatted with Arjun Desai from Cartesia, following a release of Sonic 3, a sub 49ms voice model! So, 2 interviews + tons of news, let’s dive in! (as always, show notes in the end)Hey, if you like this content, it would mean a lot if you subscribe as a paid subscriber.Open Source AIMiniMax M2: open-source agentic model at 8% of Claude’s price, 2× speed (X, Hugging Face )We kicked off our open-source segment with a banger of an announcement and a special guest. The new king of open-source LLMs is here, and it’s called MiniMax M2. We were lucky enough to have Skyler Miao, Head of Engineering at Minimax, join us live to break it all down.M2 is an agentic model built for code and complex workflows, and its performance is just staggering. It’s already ranked in the top 5 globally on the Artificial Analysis benchmark, right behind giants like OpenAI and Anthropic. But here’s the crazy part: it delivers nearly twice the speed of Claude 3.5 Sonnet at just 8% of the price. This is basically Sonnet-level performance, at home, in open source.Skylar explained that their team saw an “impossible triangle” in the market between performance, cost, and speed—you could only ever get two. Their goal with M2 was to build a model that could solve this, and they absolutely nailed it. It’s a 200B parameter Mixture-of-Experts (MoE) model, but with only 10B active parameters per inference, making it incredibly efficient.One key insight Skylar shared was about getting the best performance. M2 supports multiple APIs, but to really unlock its reasoning power, you need to use an API that passes the model’s “thinking” tokens back to it on the next turn, like the Anthropic API. Many open-source tools don’t support this yet, so it’s something to watch out for.Huge congrats to the MiniMax team on this Open Weights (MIT licensed) release, you can find the model on HF! MiniMax had quite a week, with 3 additional releases, MiniMax speech 2.6, an update to their video model Hailuo 2.3 and just after the show, they released a music 2.0 model as well! Congrats on the shipping folks! OpenAI drops gpt-oss-safeguard - first open-weight safety reasoning models for classification ( X, HF )OpenAI is back on the open weights bandwagon, with a finetune release of their previously open weighted gpt-oss models, with gpt-oss-safeguard. These models were trained exclusively to help companies build safeguarding policies to make sure their apps remains safe! With gpt-oss-safeguards 20B and 120B, OpenAI is achieving near parity with their internal safety models, and as Nisten said on the show, if anyone knows about censorship and safety, it’s OpenAI! The highlight of this release is, unlike traditional pre-trained classifiers, these models allow for updates to policy via natural language!These models will be great for businesses that want to safeguard their products in production, and I will advocate to bring these models to W&B Inference soon! A Humanoid Robot in Your Home by 2026? 1X NEO announcement ( X, Order page, Keynote )Things got really spooky when we started talking about robotics. The company 1X, which has been on our radar for a while, officially launched pre-orders for NEO, the world’s first consumer humanoid robot designed for your home. And yes, you can order one right now for $20,000, with deliveries expected in early 2026.The internet went crazy over this announcement, with folks posting receipts of getting one, other folks stoking the uncanny valley fears that Sci-fi has built into many people over the years, of the Robot uprising and talking about the privacy concerns of having a human tele-operate this Robot in your house to do chores. It can handle chores like cleaning and laundry, and for more complex tasks that it hasn’t learned yet, it uses a teleoperation system where a human “1X Expert” can pilot the robot remotely to perform the task. This is how it collects the data to learn to do these tasks autonomously in your specific home environment.The whole release is very interesting, from the “soft and quiet” approach 1X is taking, making their robot a 66lbs short king, draped in a knit sweater, to the $20K price point (effectively at loss given how much just the hands cost), the teleoperated by humans addition, to make sure the Robot learns about your unique house layout. The conversation on the show was fascinating. We talked about all the potential use cases, from having it water your plants and look after your pets while you’re on vacation to providing remote assistance for elderly relatives. Of course, there are real privacy concerns with having a telepresence device in your home, but 1X says these sessions are scheduled by you and have strict no-go zones.Here’s my prediction: by next Halloween, we’ll see videos of these NEO robots dressed up in costumes, helping out at parties. The future is officially here. Will you be getting one? If not this one, when will you think you’ll get one? OpenAI’s Grand Plan: From Recapitalization to ASIThis was by far the biggest update about the world of AI for me this week! Sam Altman was joined by Jakub Pachocki, chief scientist and Wojciech Zaremba, a co-founder, on a live stream to share an update about their corporate structure, plans for the future, and ASI goals (Artificial Superintelligence) First, the company now has a new structure: a non-profit OpenAI Foundation governs the for-profit OpenAI Group. The foundation starts with about 26% equity and has a mission to use AI for public good, including an initial $25 billion commitment to curing diseases and building an “AI Resilience” ecosystem.But the real bombshells were about their research timeline. Chief Scientist Jakub Pachocki stated that they believe deep learning systems are less than a decade away from superintelligence (ASI). He said that at this point, AGI isn’t even the right goal anymore. To get there, they’re planning to have an “AI research intern” by September 2026 and a fully autonomous AI researcher comparable to their human experts by March 2028. This is insane if you think about it. As Yam mentioned, OpenAI is already shipping at an insane speed, releasing Models and Products, Sora, Atlas, Pulse, ChatGPT app store, and this is with humans, assisted by AI. And here, they are talking about complete and fully autonomous researchers, that will be infinitely more scalable than humans, in the next 2 years. The outcomes of this are hard to imagine and are honestly mindblowing. To power all this innovation, Sam revealed they have over $1.4 trillion in obligations for compute (over 30 GW). And said even that’s not enough. Their aspiration is to build a “compute factory” capable of standing up one gigawatt of new compute per week, and he hinted they may need to “rethink their robotics strategy” to build the data centers fast enough. Does this mean OpenAI humanoid robots building factories? 🤔 Plus, don’t forget, Sam is one of the investors in Helion energy, working on power solutions like Fusion, and the above graphic has an Energy block that Sam said they will give an update on later (that’s also what he told me during Dev Day when I asked him about it). Super exciting and honestly mind-blowing stuff, Gigawats per week, fully autonomous researchers, the world is going to look way different in a few years! The Agent Labs Race: Cursor 2.0 vs. Cognition’s SWE-1.5 (X, Blog)This week also saw a major showdown in the agentic coding space. On the very same day, both Cursor and Cognition launched major updates and their own new models, signaling a new era where agent labs are training their own specialized AI.First up, Cursor 2.0 was released with a completely redesigned multi-agent interface and their new model, Composer. Composer is claimed to be four times faster than comparable models, and the new UI is built around managing a fleet of agents that can work in parallel on your codebase. It’s a clear shift from being just an IDE to a full-fledged agent platform. Look, the UI even looks like ChatGPT and no code in sight (until you switch to IDE mode) Their Composer model is also very interesting, and got a lot of folks excited, but the evaluations they shared, and the fact that they didn’t disclose if that’s a finetune of a chinese model (it likely is). Regardless, folks are saying that it’s a very good model that’s also VERY fast! Cognition own coding model - SWE 1.5 ( Blog, X, Windsurf )Then, just hours later, Cognition punched right back with SWE-1.5, their new frontier agent model that now powers Windsurf. The headline here is pure speed. Powered by Cerebras, SWE-1.5 hits a blistering 950 tokens per second—13 times faster than Sonnet 4.5—while achieving near-SOTA performance on SWE-Bench Pro. They’ve achieved this through a co-designed stack where the agent harness, inference system, and model were all built together and optimized with end-to-end reinforcement learning in real coding environments.This competition is fantastic news for all of us. We’re seeing specialized, highly-performant models being developed outside of the big labs, putting more power back in the hands of developers.This Week’s BuzzJust a few quick updates from the world of Weights & Biases and our parent company, CoreWeave.First, big news! CoreWeave announced the acquisition of Marimo, the company behind the popular open-source, reactive notebook for Python. This is another exciting step in building out the essential cloud for AI, adding powerful development tools to the stack alongside best-in-class GPU infrastructure and MLOps with Weights & Biases. Welcome to the M

Hey everyone, Alex here! Welcome... to the browser war II - the AI edition! This week we chatted in depth about ChatGPT’s new Atlas agentic browser, and the additional agentic powers Microsoft added to Edge with Copilot Mode (tho it didn’t work for me) Also this week was a kind of crazy OCR week, with more than 4 OCR models releasing, and the crown one is DeepSeek OCR, that turned the whole industry on it’s head (more later) Quite a few video updates as well, with real time lipsync from Decart, and a new update from LTX with 4k native video generation, it’s been a busy AI week for sure! Additionally, I’ve had the pleasure to talk about AI Browsing agents with Paul from BrowserBase and real time video with Kwindla Kramer from Pipecat/Daily, so make sure to tune in for those interviews, buckle up, let’s dive in! Thanks for reading ThursdAI - Recaps of the most high signal AI weekly spaces! This post is public so feel free to share it.Open Source: OCR is Not What You Think It Is (X, HF, Paper)The most important and frankly mind-bending release this week came from DeepSeek. They dropped DeepSeek-OCR, and let me tell you, this is NOT just another OCR model. The cohost were buzzing about this, and once I dug in, I understood why. This isn’t just about reading text from an image; it’s a revolutionary approach to context compression.We think that DeepSeek needed this as an internal tool, so we’re really grateful to them for open sourcing this, as they did something crazy here. They are essentially turning text into a visual representation, compressing it, and then using a tiny vision decoder to read it back with incredible accuracy. We’re talking about a compression ratio of up to 10x with 97% decoding accuracy. Even at 20x compression they are achieving 60% decoding accuracy! My head exploded live on the show when I read that. This is like the middle-out compression algorithm joke from Silicon Valley, but it’s real. As Yam pointed out, this suggests our current methods of text tokenization are far from optimal.With only 3B and ~570M active parameters, they are taking a direct stab at long context inefficiency, imagine taking 1M tokens, encoding them into 100K visual tokens, and then feeding those into a model. Since the model is tiny, it’s very cheap to run, for example, alphaXiv claimed they have OCRd’ all of the papers on ArXiv with this model for $1000, a task that would have cost $7500 using MistalOCR - as per their paper, with DeepSeek OCR, on a single H100 GPU, its possible to scan up to 200K pages! 🤯 Really innovative stuff! OCR and VLM models had quite a week, with multiple models besides DeepSeek OCR releasing, models like Liquids LFM2-VL-3B (X, HF), and the newly updated 2B and 32B of Qwen3-VL (X, Hugging Face), and AI2’s olmo-ocr 2-7B (X, HF). The Qwen models are particularly interesting, as the 2B model is a generic VLM (can also do OCR) and is close to previous weeks 4B and 8B brothers, and the newly updated 32B model outperforms GPT-5 mini and Claud 4 sonnet even! The Browser Wars are BACK: OpenAI & Microsoft Go AgenticLook, I may be aging myself here, but I remember, as a young frontend dev, having to install 5 browers at once to test them out, Chrome, Internet Explorer, Firefox, Opera etc’. That was then, and now, I have Dia, Comet, and the newly released Atlas, and, yeah, today I even installed Microsoft Edge to test their AI features! It seems like the AI boom brought with it a newly possible reason for folks to try and take a bite out of Chrome (who’s agentic features are long rumored with project mariner but are nowhere to be found/shipped yet) OpenAI’s ChatGPT Atlas: The Browser Reimagined (X, Download)OpenAI is proving that besides just models, they are a product powerhouse, stepping into categories like Shopping (with a shopify integration), app stores (with ChatGPT apps), social (with Sora2) and now... browsers! This week, they have launched their tightly integrated into ChatGPT browser called Atlas, and it’s a big release! I’ll split my review here to 2 parts, the browser features part and the agentic part. New fresh take on a chromium based browserThe tight integration into ChatGPT is everywhere in this browser, from the new tab that looks like the basic ChatGPT interaface, one line of text, to the sidebar on the left that... is the ChatGPT web sidebar with all your chats, projects, custom GPTs etc. The integration doesn’t stop there, as you have to sign in to your ChatGPT account to even use this browser (available only to MacOS users, and Pro, Plus and Nano tiers). The browser has a few neat tricks, like a special tool that allows you to search your browsing history with natural language, a-la “what were those shoes I was looking at a few days ago” will find your the tabs you browsed for shoes. A special and cool feature is called, confusingly “Cursor”, wherein you can select a text, and then click the little OpenAI logo that pops up, allowing you to ask ChatGPT for changes to that selected text (like fix typos, spruce up your writing etc). It’s surprisingly convenient to rewrite tweets or for any type of document editing. ChatGPT Atlas also stores memories about your browsing patterns, which will be additional to the ChatGPT memories it stores about you from chats, helping even more by knowing your browsing patterns, which software you prefer to use, which websites you prefer to order food from etc. This IMO is one of the hugest unlocks for folks inside the ChatGPT ecosystem, as much of a stanard persons peferences can be gleaned from their browser usage and patterns.Lastly, the “Ask ChatGPT” sidepane on the right (which can be opened with cmd+.) is really great for chatting with a webpage, or going down search rabbit holes. It receives the context of the webpage you’re looking at by default (only 1 page so far, competitors allow you to add additional tabs with @, (which is supposedly coming to ChatGPT soon) and ask... ChatGPT anything about this. Agentic SOTA? not so fastThe most important “change” to how browsers work in Atlas imo is the agentic mode. This isn’t new, we remember when ChatGPT launched thier Operator Agent back in January of this year (our coverage) and then renamed it Agent Mode and integrated into ChatGPT itself back in July. So, web browsing agents are not entirely new, what’s novel here though, is the integration into your browser, and the ability for the Atlas browser to use your logged in sessions and cookies, to pretend to be you! This... can be quite scary for some, as prompt injection attacks are getting more popular (where-in malicious a******s add hidden instructions to their website that will get the agent to do something you don’t like) but it’s also very exciting, as the agent can do much much more, without getting blocked by providers who could previously just block Agent Mode as it ran on OpenAI servers! Until today, there were 2 main Agentic browsers in the mix, Perplexity’s Comet (where you can choose which model runs the agent) and Atlas. Comet seems to be doing a little bit better on some stuff on my tests, but not by much. I have the same agentic task (go to X.com, find my bookmarks, open all links, summarize per my specific format) that I’ve been running for a while now, and Comet outdid Atlas this week on that task.Who needs agentic browsing? For some reason, most of the demos for agentic browsing are showing the same, boring-ish examples. Book some flights, collect a grocery shopping cart. I’ve tried new and different things this week, for example, letting Atlas choose and order food for me (as ChatGPT knows my pescatarian preferences, it’s better than Comet for personal stuff), and one of the longest task I’ve had an agent do yet, I asked it to complete a Compliance training I had to take at work! Mind you, this is a very complex task, even for regular people, as these compliance websites are built to not be messed with. They have video players that stop if you switch focus to some other tab, they have interactive quizes and games, drag and drop interfaces, audio buttons, to make sure you really are taking the test. I can happily report, that after 5 hours, and a few stops along the way (where I had to convince the agent to keep going), it completed this very hard task! (and now I have to take this course myself again to actualy be compliant 😅 it will probably take me 2 hours to do manually) This experiment made me think, who needs the agentic browsing features and for what? Well, for tasks that require a lot of manual steps to do the same thing over and over again, agentic browser is going to make a lot of peoples browsing a lot easier. Things like kids schedules reviewing in multiple websites, collecitng data and formatting it differently etc. Scary security implications Atlas could only finish my compliance task while being logged in as me, and ChatGPT Atlas gives a all or nothing control. You can run your agent with full access to your logged in websites (think Gmail etc) or you can essentially give it an incognito mode. This, again, due to the risk of promp injections in malicious websites being more and more prevalent. In a rare post detailing how they are thinking about this, OpenAI Chief Information Security officer offered a deep dive into their attempts to mitigate this issue (Simon Willison had a great breakdown of that information here) but that’s likely not enough, so definitely be aware when you’re running agent mode (which needs to be explicitly turned on right now by selecting Agent) This Weeks Buzz - Weights & Biases // CoreweaveWeights & Biases (now proudly part of CoreWeave) had some exciting updates. Our Fully Connected conference series is hitting Tokyo on October 30-31 and London on November 4-5—perfect for ML practitioners and AI engineers. If you’re in the area, join us for talks, networking, and deep dives into the latest. Register at Fullyconnected.com—DM me if you need a hook-up!We also collaborated with Meta and Stanfo

Hey folks, Alex here. Can you believe it’s already the middle of October? This week’s show was a special one, not just because of the mind-blowing news, but because we set a new ThursdAI record with four incredible interviews back-to-back!We had Jessica Gallegos from Google DeepMind walking us through the cinematic new features in VEO 3.1. Then we dove deep into the world of Reinforcement Learning with my new colleague Kyle Corbitt from OpenPipe. We got the scoop on Amp’s wild new ad-supported free tier from CEO Quinn Slack. And just as we were wrapping up, Swyx ( from Latent.Space , now with Cognition!) jumped on to break the news about their blazingly fast SWE-grep models. But the biggest story? An AI model from Google and Yale made a novel scientific discovery about cancer cells that was then validated in a lab. This is it, folks. This is the “let’s f*****g go” moment we’ve been waiting for. So buckle up, because this week was an absolute monster. Let’s dive in!ThursdAI - Recaps of the most high signal AI weekly spaces is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.Open Source: An AI Model Just Made a Real-World Cancer DiscoveryWe always start with open source, but this week felt different. This week, open source AI stepped out of the benchmarks and into the biology lab.Our friends at Qwen kicked things off with new 3B and 8B parameter versions of their Qwen3-VL vision model. It’s always great to see powerful models shrink down to sizes that can run on-device. What’s wild is that these small models are outperforming last generation’s giants, like the 72B Qwen2.5-VL, on a whole suite of benchmarks. The 8B model scores a 33.9 on OS World, which is incredible for an on-device agent that can actually see and click things on your screen. For comparison, that’s getting close to what we saw from Sonnet 3.7 just a few months ago. The pace is just relentless.But then, Google dropped a bombshell. A 27-billion parameter Gemma-based model they developed with Yale, called C2S-Scale, generated a completely novel hypothesis about how cancer cells behave. This wasn’t a summary of existing research; it was a new idea, something no human scientist had documented before. And here’s the kicker: researchers then took that hypothesis into a wet lab, tested it on living cells, and proved it was true.This is a monumental deal. For years, AI skeptics like Gary Marcus have said that LLMs are just stochastic parrots, that they can’t create genuinely new knowledge. This feels like the first, powerful counter-argument. Friend of the pod, Dr. Derya Unutmaz, has been on the show before saying AI is going to solve cancer, and this is the first real sign that he might be right. The researchers noted this was an “emergent capability of scale,” proving once again that as these models get bigger and are trained on more complex data—in this case, turning single-cell RNA sequences into “sentences” for the model to learn from—they unlock completely new abilities. This is AI as a true scientific collaborator. Absolutely incredible.Big Companies & APIsThe big companies weren’t sleeping this week, either. The agentic AI race is heating up, and we’re seeing huge updates across the board.Claude Haiku 4.5: Fast, Cheap Model Rivals Sonnet 4 Accuracy (X, Official blog, X)First up, Anthropic released Claude Haiku 4.5, and it is a beast. It’s a fast, cheap model that’s punching way above its weight. On the SWE-bench verified benchmark for coding, it hit 73.3%, putting it right up there with giants like GPT-5 Codex, but at a fraction of the cost and twice the speed of previous Claude models. Nisten has already been putting it through its paces and loves it for agentic workflows because it just follows instructions without getting opinionated. It seems like Anthropic has specifically tuned this one to be a workhorse for agents, and it absolutely delivers. The thing to note also is the very impressive jump in OSWorld (50.7%), which is a computer use benchmark, and at this price and speed ($1/$5 MTok input/output) is going to make computer agents much more streamlined and speedy! ChatGPT will loose restrictions; age-gating enables “adult mode” with new personality features coming (X) Sam Altman set X on fire with a thread announcing that ChatGPT will start loosening its restrictions. They’re planning to roll out an “adult mode” in December for age-verified users, potentially allowing for things like erotica. More importantly, they’re bringing back more customizable personalities, trying to recapture some of the magic of GPT-4.0 that so many people missed. It feels like they’re finally ready to treat adults like adults, letting us opt-in to R-rated conversations while keeping strong guardrails for minors. This is a welcome change, and we’ve been advocating for this for a while, and it’s a notable change from the XAI approach I covered last week. Opt in for adults with verification while taking precautions vs engagement bait in the form of a flirty animated waifu with engagement mechanics. Microsoft is making every windows 11 an AI PC with copilot voice input and agentic powers (Blog,X)And in breaking news from this morning, Microsoft announced that every Windows 11 machine is becoming an AI PC. They’re building a new Copilot agent directly into the OS that can take over and complete tasks for you. The really clever part? It runs in a secure, sandboxed desktop environment that you can watch and interact with. This solves a huge problem with agents that take over your mouse and keyboard, locking you out of your own computer. Now, you can give the agent a task and let it run in the background while you keep working. This is going to put agentic AI in front of hundreds of millions of users, and it’s a massive step towards making AI a true collaborator at the OS level.NVIDIA DGX - the tiny personal supercomputer at $4K (X, LMSYS Blog)NVIDIA finally delivered their promised AI Supercomputer, and while the excitement was in the air with Jensen hand delivering the DGX Spark to OpenAI and Elon (recreating that historical picture when Jensen hand delivered a signed DGX workstation while Elon was still affiliated with OpenAI). The workstation was sold out almost immediately. Folks from LMSys did a great deep dive into specs, all the while, folks on our feeds are saying that if you want to get the maximum possible open source LLMs inference speed, this machine is probably overpriced, compared to what you can get with an M3 Ultra Macbook with 128GB of RAM or the RTX 5090 GPU which can get you similar if not better speeds at significantly lower price points. Anthropic’s “Claude Skills”: Your AI Agent Finally Gets a Playbook (Blog)Just when we thought the week couldn’t get any more packed, Anthropic dropped “Claude Skills,” a huge upgrade that lets you give your agent custom instructions and workflows. Think of them as expertise folders you can create for specific tasks. For example, you can teach Claude your personal coding style, how to format reports for your company, or even give it a script to follow for complex data analysis.The best part is that Claude automatically detects which “Skill” is needed for a given task, so you don’t have to manually load them. This is a massive step towards making agents more reliable and personalized, moving beyond just a single custom instruction and into a library of repeatable, expert processes. It’s available now for all paid users, and it’s a feature I’ve been waiting for. Our friend Simon Willison things skills may be a bigger deal than MCPs! 🎬 Vision & Video: Veo 3.1, Sora Gets Longer, and Real-Time WorldsThe AI video space is exploding. We started with an amazing interview with Jessica Gallegos, a Senior Product Manager at Google DeepMind, all about the new Veo 3.1. This is a significant 0.1 update, not a whole new model, but the new features are game-changers for creators.The audio quality is way better, and they’ve massively improved video extensions. The model now conditions on the last second of a clip—including the audio. This means if you extend a video of someone talking, they keep talking in the same voice! This is huge, saving creators from complex lip-syncing and dubbing workflows. They also added object insertion and removal, which works on both generated and real-world video. Jessica shared an incredible story about working with director Darren Aronofsky to insert a virtual baby into a live-action film shot, something that’s ethically and practically very difficult to do on a real set. These are professional-grade tools that are becoming accessible to everyone. Definitely worth listening to the whole interview with Jessica, starting at 00:25:44I’ve played with the new VEO in Google Flow, and while I was somewhat (still) disappointed with the UI itself (it froze sometimes and didn’t play). I wasn’t able to upload my own videos to use the insert/remove features Jessica mentioned yet, but saw examples online and they looked great! Ingredients were also improved with VEO 3.1, where you can add up to 3 references, and they will be included in your video but not as first frame, the model will use them to condition the vidoe generation. Jessica clarified that if you upload sound, as in, your voice, it won’t show up in the model as your voice yet, but maybe they will add this in the future (at least this was my feedback to her). SORA 2 extends video gen to 15s for all, 25 seconds to pro users with a new storyboard Not to be outdone, OpenAI pushed a bit of an update for Sora. All users can now generate up to 15-second clips (up from 8-10), and Pro users can go up to 25 seconds using a new storyboard feature. I’ve been playing with it, and while the new scene-based workflow is powerful, I’ve noticed the quality can start to degrade significantly in the final seconds of a longer generation (posted my experiments here) as you can see. The last few shot so

Hey everyone, Alex here 👋We’re deep in the post-reality era now. Between Sora2, the latest waves of video models, and “is-that-person-real” cameos, it’s getting genuinely hard to trust what we see. Case in point: I recorded a short clip with (the real) Sam Altman this week and a bunch of friends thought I faked it with Sora-style tooling. Someone even added a fake Sora watermark just to mess with people. Welcome to 2025.This week’s episode and this write-up focus on a few big arcs we’re all living through at once: OpenAI’s Dev Day and the beginning of the agent-app platform inside ChatGPT, a bizarre and exciting split-screen in model scaling where a 7M recursive model from Samsung is suddenly competitive on reasoning puzzles while inclusionAI is shipping a trillion-parameter mixture-of-reasoners, and Grok’s image-to-video now does audio and pushes the line on… taste. We also dove into practical evals for coding agents with Eric Provencher from Repo Prompt, and I’ve got big news from my day job world: W&B + CoreWeave launched Serverless RL, so training and deploying RL agents at scale is now one API call away.Let’s get into it.OpenAI’s 3rd Dev Day - Live Coverage + exclusive interviewsThis is the third Dev Day that I got to attend in person, covering this for ThursdAI (2023, 2024), and this one was the best by far! The production quality of their events rises every year, and this year they’ve opened up the conference to >1500 people, had 3 main launches and a lot of ways to interact with the OpenAI folks! I’ve also gotten an exclusive chance to sit in on a fireside chat with Sam Altman and Greg Brokman (snippets of which I’ve included in the podcast, starting 01:15:00 and I got to ask Sam a few questions after that as well. Event Ambiance and VibesOpenAI folks outdid themselves with this event, the live demos were quite incredible, the location (Fort Mason), Food and just the whole thing was on point. The event concluded with a 1x1 Sam and Jony Ive chat that I hope will be published on YT sometime, because it was very insightful. By far the best reason to go to this event in person is meeting folks and networking, both OpenAI employees, and AI Engineers who use their products. It’s 1 day a year, when OpenAI makes all their employees who attend, Developer Experience folks, as you can and are encouraged to, interact with them, ask questions, give feedback and it’s honestly great! I really enjoy meeting folks at this event and consider this to be a very high signal network, and was honored to have quite a few ThursdAI listeners among the participants and OpenAI folk! If you’re reading this, thank you for your patronage 🫡 Launches and ShipsOpenAI also shipped, and shipped a LOT! Sam was up on Keynote with 3 main pillars, which we’ll break down 1 by 1. ChatGPT Apps, AgentKit (+ agent builder) and Codex/New APIsCodex & New APIsCodex has gotten General Availability, but we’ve been using it all this time so we don’t really care, what we do care about is the new slack integration and the new Codex SDK, which means you can now directly inject Codex agency into your app. This flew a bit over people’s heads, but Romain Huet, VP of DevEx at OpenAI demoed on stage how his mobile app now has a Codex tab, where he can ask Codex to make changes to the app at runtime! It was quite crazy! ChatGPT Apps + AppsSDKThis was maybe the most visual and most surprising release, since they’ve tried to be an appstore before (plugins, customGPTs). But this time, it seems like, based on top of MCP, ChatGPT is going to become a full blown Appstore for 800+ million weekly active ChatGPT users as well. Some of the examples they have showed included Spotify and Zillow, where just by typing in “Spotify” in chatGPT, you will have an interactive app with it’s own UI, right inside of ChatGPT. So you could ask it to create a playlist for you based on your history, or ask Zillow to find homes in an area under a certain $$ amount.The most impressive thing, is that those are only launch partners, everyone can (technically) build a ChatGPT app with their AppsSDK that’s built on top of... the MCP (model context protocol) spec! The main question remains about discoverability, this is where Plugins and CustomGPTs (previous attempts to create apps within ChatGPT have failed), and when I asked him about it, Sam basically said “we’ll iterate and get it right” (starting 01:17:00). So it remains to be seen if folks really need their ChatGPT as yet another Appstore. AgentKit, AgentBuilder and ChatKit2025 is the year of agents, and besides launching quite a few of their own, OpenAI will not let you, build and host smart agents that can use tools, on their platform. Supposedly, with AgentBuilder, building agents is just dragging a few nodes around, prompting and connecting them. They had a great demo on stage where with less than 8 minutes, they’ve build an agent to interact with the DevDay website.It’s also great to see how greatly does OpenAI adapt the MCP spec, as this too, is powered by MCP, as in, any external connection you want to give your agent, must happen with an MCP server. Agents for the masses is maybe not quite there yetIn reality though, things are not so easy. Agents require more than just a nice drag & drop interface, they require knowledge, iteration, constant evaluation (which they’ve also added, kudos!) and eventually, customized agents need code. I spent an hour trying it out yesterday, building an agent to search the ThursdAI archives. The experience was a mixed bag. The AI-native features are incredibly cool. For instance, you can just describe the JSON schema you want as an output, and it generates it for you. The widget builder is also impressive, allowing you to create custom UI components for your agent’s responses.However, I also ran into the harsh realities of agent building. My agent’s web browsing tool failed because Substack seems to be blocking OpenAI’s crawlers, forcing me to fall back on the old-school RAG approach of uploading our entire archive to a vector store. And while the built-in evaluation and tracing tools are a great idea, they were buggy and failed to help me debug the error. It’s a powerful tool, but it also highlights that building a reliable agent is an iterative, often frustrating process that a nice UI alone can’t solve. It’s not just about the infrastructure; it’s about wrestling with a stochastic machine until it behaves.But to get started with something simple, they have definitely pushed the envelope on what is possible without coding. OpenAI also dropped a few key API updates:* GPT-5-Pro is now available via API. It’s incredibly powerful but also incredibly expensive. As Eric mentioned, you’re not going to be running agentic loops with it, but it’s perfect for a high-stakes initial planning step where you need an “expert opinion.”* SORA 2 is also in the API, allowing developers to integrate their state-of-the-art video generation model into their own apps. The API can access the 15-second “Pro” model but doesn’t support the “Cameo” feature for now.* Realtime-mini is a game-changer for voice AI. It’s a new, ultra-fast speech-to-speech model that’s 80% cheaper than the original Realtime API. This massive price drop removes one of the biggest barriers to building truly conversational, low-latency voice agents.My Chat with Sam & Greg - On Power, Responsibility, and EnergyAfter the announcements, I’ve got to sit in a fireside chat with Sam Altman and Greg Brockman and ask some questions. Here’s what stood out:When I asked about the energy requirements for their massive compute plans (remember the $500B Stargate deal?), Sam said they’d have announcements about Helion (his fusion investment) soon but weren’t ready to talk about it. Then someone from Semi Analysis told me most power will come from... generator trucks. 15-megawatt generator trucks that just drive up to data centers. We’re literally going to power AGI with diesel trucks!On responsibility, when I brought up the glazing incident and asked how they deal with being in the lives of 800+ million people weekly, Sam’s response was sobering: “This is not the excitement of ‘oh we’re building something important.’ This is just the stress of the responsibility... The fact that 10% of the world is talking to kind of one brain is a strange thing and there’s a lot of responsibility.”Greg added something profound: “AI is far more surprising than I anticipated... The deep nuance of how these problems contact reality is something that I think no one had anticipated.”This Week’s Buzz: RL X-mas came early with Serverless RL! (X, Blog)Big news from our side of the world! About a month ago, the incredible OpenPipe team joined us at Weights & Biases and CoreWeave. They are absolute wizards when it comes to fine-tuning and Reinforcement Learning (RL), and they wasted no time combining their expertise with CoreWeave’s massive infrastructure.This week, they launched Serverless RL, a managed reinforcement learning service that completely abstracts away the infrastructure nightmare that usually comes with RL. It automatically scales your training and inference compute, integrates with W&B Inference for instant deployment, and simplifies the creation of reward functions and verifiers. RL is what turns a good model into a great model for a specific task, often with surprisingly little data. This new service massively lowers the barrier to entry, and I’m so excited to see what people build with it. We’ll be doing a deeper dive on this soon but please check out the Colab Notebook to get a taste of what AutoRL is like! Open SourceWhile OpenAI was holding its big event, the open-source community was busy dropping bombshells of its own.Samsung’s TRM: Is This 7M Parameter Model... Magic? (X, Blog, arXiv)This was the release that had everyone’s jaws on the floor. A single researcher from the Samsung AI Lab in Montreal released a paper on a Tiny Recursive Model (TRM). Get this: it’s a 7