The AI Weekly · Week of November 2, 2026
Frontier leadership flips as world models, agents, and open weights test the old order
- Anthropic's IPO Filing Reveals a Six-Month Model Obsolescence MachineArticle — "Anthropic's IPO Filing Reveals a Six-Month Model Obsolescence Machine": Eighty percent of Anthropic's revenue comes from models less than six months old. Read that sentence twice, because in any other software…
- World Models Emerge as the Post-LLM Paradigm, from Westlake to Anthropic's $6B BetOn the last day of October, in the auditorium of Westlake University's Yungu campus, roughly ninety industry R&D chiefs and investors gathered to hear about world models — and the first order of business was explaining w…
- Gemini 4 Carbon Tops the Frontier Charts as Access Stays LimitedGoogle spent much of 2026 playing catch-up in the AI race. A month and two frontier models later, it is sitting at the top of the charts again — sort of.
- DeepSeek Open-Sources Super-Agent Architecture as Agent Liability Defaults to UsersNobody @-mentions the agent. DeepSeek's own writeup sells the scene anyway: a bug surfaces in the team group chat, the agent reportedly reproduces it in the background through speculative execution, and if the fix turns …
- Mistral Ships 1T-Parameter Open Weights Trained on European GPUsArticle — "Mistral Ships 1T-Parameter Open Weights Trained on European GPUs": Last week, the open-weight cost shock was a question: can models released outside the US and China actually hold prices down? This week, Mistr…
Threads we are following
- Whether Anthropic's S-1 disclosure of decelerating sequential growth and undisclosed gross margins will force other frontier labs to reveal their own unit economics before going public.
- Whether a universal training unit for physical data can be found, enabling consolidation of world-model architectures.
- Whether Google actually opens Gemini 4 Carbon to general access, and how long its benchmark lead holds once OpenAI and Anthropic ship their next frontier models.
- Whether the first serious agent incident — a lifted credential, an unapproved merge, or an overread conversation — forces labs to ship liability architecture instead of sysadmin suggestions.
- Whether independent re-runs on the public weights confirm Mistral's cyber-defense scores and the reported token-efficiency advantage over closed and Chinese open-weight rivals.
One issue a week. Free.
sift. reads hundreds of sources so you don't have to, and sends one issue you can actually finish. No daily noise. Unsubscribe anytime.