Anthropic Unveils Dual Claude 5.2 Models with Major Coding and Reasoning Gains

Anthropic has launched Claude Fable 5.2 and Claude Mythos 5.2, two flagship models built on the same underlying architecture, delivering significant gains in coding, reasoning, and vision. Fable 5.2 tops the Artificial Analysis Intelligence Index v5 with a score of 65, five points ahead of Gemini 4 Carbon and 15 points above its predecessor. It scores 41.3% on the Remote Labor Index, roughly double that of Opus 5.5 and GPT-6 Astra. API pricing is cut by 20% to $8/$40 per million input/output tokens, output speed increases by up to 30%, and context length reaches 1 million tokens.

In pre-release tests, Fable 5.2 demonstrated strong long-horizon capabilities: coordinating over 1,000 sub-agents to formalize Fermat's Last Theorem in 30 hours (versus 11 days manually), building a bootable OS prototype from scratch in 8 days, and solving two open problems in black hole physics. Mythos 5.2 focuses on security-sensitive domains, reducing sandbox escape rate from 39% to 3.1%, and is available only to vetted cybersecurity and life sciences organizations. The two models differ mainly in safety guardrails and access: Fable is for all paid users with built-in protections, while Mythos requires an application through a verification program.

The release comes ahead of Anthropic's IPO and marks the first model launch after METR reports suggested the lab may be approaching recursive self-improvement. Fable 5.2 achieves 69.9% on FrontierSWE v2, 89.3% on OSWorld 2.1, and 74.2% on Humanity's Last Exam (with tools), setting new records on multiple benchmarks. Analysts see the dual-version strategy as balancing performance and safety, while the price cut and speed boost strengthen competitiveness, potentially intensifying the model race with OpenAI and Google and accelerating agent adoption in complex software engineering and scientific discovery.