Following gold-medal-level performance from our AI models across five competitions in mathematics, physics, and chemistry, we asked a harder question: can AI contribute when a problem is genuinely open and without an existing solution path? Over the past several months, mathematicians worked with Muse Spark 1.1 and Muse Spark 1.2 in Thinking Mode through the regular meta.ai chat interface, with no custom research scaffold, to find solutions to such problems. Our goal wasn't to mass-produce papers, but empower researchers. Every collaboration followed the same principles: mathematicians guided the research, a second group of mathematicians then reviewed the work, each paper marks which passages were drafted primarily by humans or AI, and each credits the prior research it builds on. Where other teams independently announced solutions to the same problems, we acknowledge their work as well. Today, we're sharing six papers from that collaboration. 🧵👇

We need at least 50x more compute. At least.

Only in SF😭😭😭

Half a million downloads in a month. Today, our open source family takes another step forward. Thank you for the incredible support behind our first-generation models. We’re excited to introduce TwIL-LM3-Pro. At just 3.6 billion parameters, it brings powerful reasoning to everyday computers, with quantized builds that run locally. No cloud required. In our evaluation: Formal logic: Highest recorded headline score among the small models compared—beating China’s VibeThinker-3B by 35% and Qwen3.5-4B by 24%, and Liquid AI’s LFM2.5-8B-A1B by 47%. Broader reasoning: 95% on SVAMP and 64.1% on MuSR, the highest recorded scores among the small models compared. BIG-Bench Hard’s logic subset: 95.4%, compared with VibeThinker-3B’s 61.1%. We believe AI is entering a post-training era. The advantage will increasingly belong to companies with the best pipelines and those that can produce capable, personalized intelligence faster and more efficiently, then put it on devices people already own. That’s what we’re building at webAI. And we’re only beginning to share what’s coming out of our lab. Coming soon: Meridian, our family of frontier-class models built to run on device. Our most advanced models will be available through the @thewebAI application. Join the waitlist as we expand access. Proudly built in Austin, Texas. 🇺🇸

1/ Excited to release The ultimate guide to multi-harness RL The same model behaves differently in every agent harness. So we built an open way to train any model with RL on any task set, inside the harnesses people actually use, like Claude Code, Codex, and OpenCode, without changing a single line of harness or training code. Trained across four harnesses, LFM2.5-2.6B went from 42% to 54% with 31% fewer tool calls. 🧵

Small models are getting really good at reasoning. It's exciting because SLMs can unlock so much at the harness layer. TwIL-LM3-Pro from @thewebAI has 3.6B parameters and runs locally on everyday computers. It scores 95.4 on BIG-Bench Hard, well ahead of Qwen3-8B at 63.7. I like their post-training recipe. They fine-tune on formal logic, merge the weights back toward the base model, and then run RL against a programmatic verifier. Logic scores go up, and general reasoning holds steady. Great to see more of this work released as open source.

holy

Interesting study. But let's face it, soon enough AI's will do the vast majority of inventing. 10's of thousands per year. So fast that patents won't be relevant.

Before we get space-based computing at scale, sea-based computing makes so much sense! Acceleration on the seas. 🌊/acc

California Gov. Gavin Newsom signed an executive order today (30 sep) declaring that Artificial Intelligence will be called "Artificial Intelligence" in California. > Employers can't use AI to read workers' emotions. AB 1883 bans AI surveillance tools that recognize or predict an employee's emotional state, or that collect neural data, such as readings from brain or nervous system sensors. > AI can't make firing decisions alone, and AI layoffs must be disclosed. Employers can't rely only on AI for discipline or termination. They must also tell workers if a mass layoff, relocation, or termination is caused by an AI system. > The wider package goes beyond the workplace. It protects doctors' own judgment when clinical AI tools are used, bans deleting AI watermarks, strengthens likeness protections against deepfakes, and requires gene synthesis companies to verify their customers and screen what they ship.

The frontier labs (Claude + ChatGPT + Grok + Gemini) have been log-monotonically plowing into the enterprise super intelligence market. People often ask what inning we are in? On this definition of market size, AI software is roughly 0.1% penetrated into the enterprise; the first pitch has only just been thrown and is not even a quarter of the way to the catcher's mitt. Adjusting for the realized rate of market penetration of the frontier labs, and assuming they follow a traditional diffusion curve, some $30 trillion in revenue could be up for grabs by 2030. Even if the rate of diffusion were to slow to 2/3rds that realized thus far, $10 trillion in revenue could be at stake for the AI frontier labs by 2030.

This is the NX ONEPOD, a 100kW micro data centre. A contenarised 5.8 TB of aggregated GPU memory with 32x NVIDIA B200 GPUs. Fully plug and play and backed up with batteries for 24h operations and reducing netwrok loads. This could be built for around 3M$, built and deployed in 5 months and could support approximately 50,000–150,000 people whose daily digital activity is extensively AI-assisted.

Humans are the minority user of AI Agents burn nearly 5x the tokens people do, up 14x since February More charts in State of Markets II: a16z.news/p/state-of-markets…