🔥 Diffusion LMs promise what autoregressive models can’t: fixing their mistakes. But new ones often change the noise and the sampler at once. Which matters? We tested it on the hypersphere. Noise: 0.8 points. Sampler: 25. We introduce Spherical Brownian Diffusion 🌐 👇🧵

Introducing Fern, Mirendil’s first model: pretrained from scratch, with up to 10M tokens of context! That’s ~0.5M lines of code (roughly Megatron-LM scale) or days of an agent’s work, all in context at once. Its first version, Fern 0.1, had a short round of post-training 🚢

Hot take: the community keeps asking whether diffusion LLMs can beat AR. That's the wrong question. Diffusion is a sampling method, and it can sit on top of an AR model to make it faster instead of replacing it. Speculative decoding already speeds up AR, but it has two long-standing problems. Training a drafter takes a long time, and you have to build a complete pretraining corpus for it. Even then, token acceptance drops sharply on out-of-domain data. Uno @ssahoo_ offers an elegant way out: it decouples sampling training from capability training. There's no separate drafter. The model itself, plus one LoRA, drafts several tokens in a single forward pass, and the LoRA learns to match the model's own next-token distribution, so there's no drafter corpus to build and no labels to collect. We applied this to the Looped Transformer Nanbeige4.2-3B: 1 hour of LoRA training gives a lossless 2× speedup on agentic task. We haven't optimized anything yet, and we believe the acceleration can go much further. The takeaway: diffusion doesn't need to beat AR. Any model that can score its next token can get a fast diffusion mode in an hour. Page: fengsxy.github.io/LoopUno Lora Adapters: huggingface.co/liddlefish/Lo…

Anthropic is now recommending that companies REMOVE unnecessary subagents from their AI systems. This is a pretty significant shift in agent architecture, with their team finding that a single agent equipped with skills can frequently outperform complex multi-agent systems, and one customer seeing roughly 2x faster execution. The emerging rules of thumb: → Skills: Teach the agent HOW to do something, including procedures, expertise, and instructions. → Tools / MCP: Give the agent the ability to DO things through APIs, actions, and live data. → Subagents: Use for parallel work, isolated context, or tasks a single agent struggles with. The interesting part is progressive disclosure. Instead of spawning specialized agents or stuffing everything into a giant prompt, one agent discovers and loads skills and tools as needed. Their advice is to fold existing subagents into the main loop, turn their instructions into skills, expose their actions as tools, and run evals to see if performance holds up. As models have gotten smarter, improving agent architecture is becoming less about what you can add and more about what you can delete. claude.com/resources/article…

🚨 New paper! What if LLMs could reason like we do when solving Sudoku: keep a few candidates per cell and iteratively narrow them down? Introducing Binary Discrete Diffusion for language and reasoning tasks: it denoises sets of candidate tokens using binary representations! 🧵

Introducing Kimi K3.1 Pro: our most capable 3T-class open-weight model. 🔹 1M-token context, 2.8T total / 104B active params (MoE), Multimodal. 🔹 Full-project coding: games in Unreal, Unity & Godot. 🔹 Excels at parametric 3D modeling in Blender & CAD. 🔹 More reliable at finance work: Accurate modeling with hallucination down by 65%. 🔹 Matching the strongest proprietary models across coding, agentic and knowledge benchmarks. Kimi K3.1 Pro and K3.1 Pro Highspeed (2–3× faster) are now live on Kimi.ai, Kimi Work, Kimi Code, and the Kimi API. Open weights by Nov 4, 2026. 🔗 API: platform.kimi.ai 🔗 Tech blog: kimi.ai/blog/kimi-k3-1-pro

2 months ago, I cold emailed Joe Gebbia, co-founder of Airbnb. he actually replied haha. $100B company btw. didn't get the meeting, but learned something. your cold email doesn't need to be perfect. just keep it short, make your ask clear, and hit send. sometimes even a rejection is a win :)

1/8 New paper with Ivan Ilin: "Local Work, Global Progress: A Convergence Theory of AI Agent Collaboration". We all know that teams of LLM agents can tackle big tasks. However, their collaboration protocols lack guarantees. We model collaboration as block optimization and, under suitable assumptions, prove convergence. arxiv.org/abs/2610.28286

Off-policy RL is difficult; runs can suddenly collapse because of async RL staleness or numerical mismatch. StableReinforce is a new method on Tinker that keeps your RL runs stable, even when learning drifts from sampling. tinker-docs.thinkingmachines…

We’re opening a research preview of Claude Duet, a new app where Claude works as a proactive agent. It comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf.

We’ve developed the world’s most complete megakernel framework. It can perform entire training runs, serve inference, play video games, and even do all three simultaneously in a single kernel.

I’ve figured out the best way to design with Claude: Have it build a tiny webapp that shows every component before and after, with approve/deny/feedback buttons. Then add one button that sends Claude all your feedback at once.

1/4 Today, we’re releasing Mercury 3. Mercury 3 is by far our most powerful model and the most intelligent diffusion LLM ever released. It outperforms GPT-6 Luna and Haiku 5.5 on quality while running at 1000+ tok/sec.

pstack v0.15.27 ships with a new skill /create-formal-verification-skill! it's still in beta, and similar to /create-verification-skill, this is a meta-skill that creates a new skill and formal verification suite based on the characteristics you want to verify. for example, you may want to model state machines and systems with concurrency using TLA+, or use kani in rust to check for undefined behavior. if you're new to formal verification, think of it as mathematically proving properties and invariants (something that must always be true) or "laws" of your system. usually, this means modeling your programs in a separate language like with TLA+ and then running the solver on that spec. newer approaches like with kani and @bendlang take a different approach where there is no separate language: you write kani in rust, and bend in bend (it's also a programming language, and the proofs are written as laws and expressed in its type system). historically formal verification never reached mainstream status because of the difficulty in modeling and writing proofs. it's quite difficult to model your business requirements and keep them in sync! but i expect that we will see more innovation in formal verification in the future as being able to formally verifiy that the code does what you intended gives your agents superpowers in writing correct software. in the meantime though, i've been experimenting with modeling our code with TLA+ and have already found several bugs with the skill created by /create-formal-verification-skill. try it out and let me know how it goes!

Generative UI #2 - Merge conflicts 🚫 ✨

Today we’re releasing Muse Spark Open, the most capable US open-weight model to date. Built for coding and agentic work, it excels at long-horizon tasks, complex software engineering, and multi-step reasoning. Key Model Specs: → 816B total / 42B active parameters → 1M token context window → Natively multimodal across text, image, video, audio