Anthropic has officially filed to go public. In this week's AI news, Jeff Keltner and Annie Delgado break down the Anthropic S-1: $16 billion in revenue in the first half of the year, almost $1 billion in operating profit, and an $80 billion loss that is mostly an accounting effect of its rising valuation. They also cover the $800 billion in compute commitments, the roughly $2 trillion valuation, and why OpenAI is holding back its own IPO.
Also in this episode:
Can you trust AI benchmarks? How bench maxing works, and what the new leaderboard changes for open models
New model releases: Gemini 4 Carbon, Kimi K3.1 Pro and GPT 6.1 Terra
OpenAI and Anthropic both claim to have disproved the exponential time hypothesis on the same day
The PHP-to-shell vulnerability and the delay before CISA listed it
The new executive order that replaces "AI" with "superintelligence" in some official language
Google's Genie 4 world model: navigable 3D worlds generated from a prompt, with real-time audio and robot-training uses
Subscribe for weekly AI news. Get the full readout in the newsletter at what theai.fm.