Good morning. Anthropic’s IPO roadshow has arrived with a number attached — $1.4 trillion — and Google picked today to drop its answer to Fable 5.2. Meanwhile, New York’s AI safety law is now live, Amazon quietly admits it was using Sonnet to sort IT tickets, and the Jensen Huang story keeps getting worse.
Anthropic prices its IPO at a $1.4 trillion valuation. The company filed updated S-1 docs setting a $70–$78 share range, with trading expected on Nasdaq under “ANT” around November 19. That’s below the rumored $2 trillion target but still ahead of OpenAI’s $1.25 trillion mark, and it comes on top of $81 billion in annualized run-rate revenue as of October. TechCrunch has the filing details, while The Verge attributes the softer-than-expected pricing to a mix of safety debates, midterm results that didn’t go AI’s way, and the ongoing Huang mess.
Each of the seven co-founders will cash out about $250M. Per TechCrunch’s read of the S-1, all seven founders are selling exactly 3,378,378 shares apiece while collectively retaining 51.1% voting control via supervoting Class F stock. Reed Hastings has signaled he’ll buy $300 million through his donor-advised fund. Google and Nvidia are each putting in $10 billion via private placement — a notable detail given Google just launched a direct competitor today.
Google ships Gemini 4 Carbon. DeepMind’s new frontier model is now broadly available via API, AI Studio, and the Gemini app, priced at $2/$10 per million tokens — roughly a fifth of what GPT-6 Astra charges and a quarter of Fable 5.2. Google claims #1 rankings on CursorBench 4.0, Perplexity’s WANDR, and Zapier’s AutomationBench at over 2x lower cost per task. DeepMind’s blog post has the benchmarks; The Verge notes a preview “/swarm” multi-agent feature in the Antigravity IDE that mirrors what Anthropic is doing with Fable subagents.
The Fable 5.2 reception is coming into focus. Early users on Hacker News are impressed but calibrated. One team reported a 36x speedup on a 600k-line Fortran codebase for computational materials science and plans to open-source the work. Others are picking apart the benchmark charts — specifically why Anthropic stacked Fable xHigh against Carbon Med and left Carbon off the single cyber comparison. Ethan Mollick’s early review leans positive but notes Opus 5.5 is still good enough that Fable FOMO isn’t what it was last cycle.
DeepMind also shipped AlphaProof Code. The new agent, internally called CodeMender, applies formal verification to real codebases by generating mathematical proofs of correctness, with cryptographic libraries as the initial target. HN commenters were skeptical of the claim that a non-trivial human-written library verified to zero bugs — one noted it’s easy for an LLM writing Lean proofs to cheat by weakening the spec. The optimistic read, from another commenter: in two or three years it may seem mad we ever ran unverified software, the way plaintext HTTP seems mad now. Google’s writeup has the specifics.
New York’s RAISE Act is now in force. Frontier developers with $500M+ in revenue — OpenAI, Anthropic, Google — must register by January 1, publish safety frameworks, report critical incidents within 72 hours, and issue transparency reports, with fines up to $3 million for repeat violations. The state also seeded a $55 million AI safety research center at SUNY Binghamton. TechCrunch’s coverage puts it alongside California and Illinois as part of the state-level push filling the federal vacuum.
Amazon replaced Sonnet 4.6 with a fine-tuned Gemma 4 2B for ticket classification. The switch to a purpose-built “Strands Decider” model cut costs by 98% while matching accuracy, per the Strands blog. The HN reaction was less “nice optimization” and more “why was Amazon using Sonnet for this in the first place?” One commenter recalled building a ticket router on SVMs six years ago that trained in minutes and classified in milliseconds. Another put it plainly: LLMs became a hammer, and the field forgot a decade of foundational ML work exists.
The Jensen Huang story has a timeline now. TechCrunch has published a chronology of the affair with Alice Huang, the Megaspeed executive U.S. officials believe fed intelligence to Beijing. The alleged network reportedly acquired billions in Nvidia chips including export-banned models for Chinese entities, and Huang’s own legal team is said to have intervened in a related government probe. Trump has publicly defended him; Huang has pulled from public speaking.
Expect a quieter day tomorrow — or whatever today’s equivalent of that phrase is, given Anthropic starts its investor roadshow and Google’s Carbon rollout is still landing in developers’ hands.