General
- OpenAI: Research acceleration: The view inside OpenAI.
- Anthropic: Measurements for understanding the pace of AI development inside frontier labs.
- Epoch AI: AI is getting cheaper more quickly than any other transformative tech in history. The plunging price of thought.
Safety
- A renewed interest in coordination among labs to pace the frontier:
- Dario Amodei: We Must Pace the Frontier.
- Sam Altman concurs: The world deserves confidence that American companies developing increasingly capable AI will act responsibly, especially as the trajectory of progress has steepened.
- Demis Hassabis concurs: A Framework for Frontier AI and the Dawning of a New Age.
- Ryan Greenblatt: Latent reasoning architectures would undermine CoT, our strongest oversight tool.
LLM
- Anthropic announce Claude Fable 5.1 and Mythos 5.1.
- OpenAI reveals information about their upcoming Astra model. Astra apparently exploits a recurrent neural pathway (perhaps like a looped transformer), meaning more of the reasoning is happening in opaque neuralese compared to the English chain-of-thought. This makes monitorability more challenging.
- Google release Gemini 3.8 Flash.
- Meta release Muse Spark 1.3.
- OpenAI announce GPT-6 Astra. They claim “Welcome to the AGI era”. Impressive benchmark scores, including >98% on ARC-AGI-3 (previous record <8%).
- Google announce Gemini 3.8 Live and 3.8 Live Extended Thinking.
- Anthropic release Claude Opus 5.5.
- OpenAI release GPT-6 Sol and Luna.
- Anthropic release Claude Sonnet 5.5.
- OpenAI release GPT-6.1 Sol.
Agents
Image Synthesis
- OpenAI release ChatGPT Images v2.5.
Video
- Google adds agentic video understanding to Gemini.
World Synthesis
- World Labs announces Atlas, a world model (3D/4D, video, robotics, etc.). Impressive 3D reconstructions, 4D video, time freeze, etc.
- Runway ML describe a bit more about their General World Model (GWM 2).
Science
- Google announce WeatherNext 3, global weather model.
- Anthropic: Formalizing Fermat’s Last Theorem (Lean, reaction).
- Google report: AlphaGenome Atlas: A predictive map of every possible DNA letter change in the human genome (try using).
- AI solutions to Navier-Stokes Millennium Prize Problem
- Levent Alpöge and Tristan Buckmaster solution obtained using OpenAI and Anthropic models.
- OpenAI: On the Navier–Stokes Millennium Prize Problem; paper: FINITE TIME BLOWUP FOR NAVIER–STOKES. Solution obtained by a group of agents using next-generation model (better than GPT-6 Astra). This involved ~10,000 agents working for 88 hours, generating 2.7 million messages and using approximately 130 billion output tokens.
- Periodic Labs: Building Labs that Learn. Nature Is Our Learning Environment (trained Periodic Neon to do XRD). AI Infrastructure at Periodic.
- Anthropic: Yes, Claude can do Nine Loops.
Brain
- 5 amazing visuals show how the male fruit fly’s brain map is advancing neuroscience A years-long project by HHMI Janelia, Google Research, and collaborators has built the first complete brain map for a male fruit fly, a key model organism in science.
Cars
- Waymo Safety Impact. Safer than human drivers.
Robots
- Reward AI announce: OM-1: Frontier Robot Intelligence, Learned Firsthand from Humans. Learned from humans (not teleop), generalizes to any robot form factor.
- Odyssey announce Odyssey-3, a foundation world model.
- Figure announce: Helix 2.5: Zero-Shot 30-Home Generalization.