Claude Fable 5.2 formalized Fermat's Last Theorem in two days, Anthropic releases two flagship models at once

Anthropic released two flagship models, Claude Fable 5.2 and Claude Mythos 5.2, sharing the same underlying model but differing in safeguards, with Fable 5.2 open to all users and Mythos 5.2 gated behind Anthropic's expanded life-sciences and cybersecurity verification programs. The models are priced at $8 per million input tokens and $40 per million output tokens, with an optional Fast Mode at $16/$80 that boosts generation speed up to 2.5x, and they generate up to 30% faster than the 5.1 versions. Anthropic said Fable 5.2 formalized Fermat's Last Theorem in 30 hours and topped benchmarks including FrontierCode 2.0 (49.1% vs. GPT-6 Astra's 42.8%), FrontierSWE v2 (69.9% vs. 65.5%), RoboDojo Sim (64.9%, up 37.3 points from Fable 5.1's 27.6%) and OSWorld 2.1 (89.3%), though Opus 5.5 retained a slight edge on AA-Briefcase with an Elo of 1,800 versus 1,785. Mythos 5.2 found 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1 on Anthropic's internal CVE Rediscovery eval, with one security firm reporting that one in six confirmed flaws it found had been missed by every other frontier model tested. The launch intensifies the flagship-model race with OpenAI's GPT-6 Astra, and the key things to watch are enterprise adoption of the pricier Fast Mode, uptake of the gated Mythos access programs, and whether Anthropic's claimed cost efficiency at high reasoning effort holds up under independent testing.