Anthropic launches Claude Fable 5.2 and Mythos 5.2, days before setting its IPO price
Follow-up to Sunday Nov 8: METR: Anthropic's AI now improves twice as fast, and helps crack a quantum puzzleRemember Sunday's token on METR? The safety group said Anthropic's models now improve about twice as fast as a year ago, with AI helping to build AI. On Monday, Anthropic released two new models. TechCrunch's headline says they arrive "amid RSI concerns". RSI means recursive self-improvement: AI that helps make better AI.
Anthropic says Claude Fable 5.2 and Claude Mythos 5.2 are better at coding, reasoning and vision, and faster than the 5.1 versions. Fable 5.2 targets coding and agent tasks, and it scored 65 on Artificial Analysis, the highest so far. Mythos 5.2 targets cybersecurity and life-sciences research, and only organizations approved by Anthropic can use it. Anthropic also says Fable "solved two open problems in the physics of black holes". The launch was the top story on Hacker News, with 977 points on one post and 803 comments on another.
The timing matters. The Wall Street Journal reports Anthropic is expected to set its IPO valuation range this week. A Reddit post also claims Anthropic has already shown a model stronger than Fable 5.2. That is unconfirmed.
New frontier models now arrive every few weeks, and a stock market sale adds pressure to keep shipping. Keeping Mythos for approved groups is a good sign. But the vetting, the testing and the safety claims all come from the lab itself, and nobody outside checks them before launch.
