Anthropic releases Claude Fable 5.2 generally and gated Claude Mythos 5.2 at lower API prices

Anthropic released Claude Fable 5.2 for general access alongside Claude Mythos 5.2 for verified security and life-sciences teams on 9 November 2026. Unlike separate model sizes, both variants share the same underlying weights and 1M-token context window: Fable 5.2 applies automated classifiers and refusal safeguards against dual-use cyber and biological misuse, while Mythos 5.2 removes those filters for organisations approved through Anthropic’s Cyber Verification Program and Life Sciences Verification Program.

Access, speed and pricing

Claude Fable 5.2 is available across the Claude Platform API, Claude Code, Claude Pro via usage credits, and Claude Max and Ultra plans up to 50% of a subscriber’s weekly usage allowance. Anthropic states that the 5.2 generation runs up to 30% faster than 5.1 while lowering standard API pricing from $10 and $50 to $8 per million input tokens and $40 per million output tokens. Prompt-cache reads cost $0.20 per million tokens, with cache writes billed at $10 for five-minute retention or $16 for one-hour retention. An optional fast mode increases output speed by up to 2.5 times at $16 per million input tokens and $80 per million output tokens.

Scientific and security evaluations

In prerelease evaluations reported by Anthropic, the model designed three working electronic devices from scratch, solved two open black-hole physics problems, and formalised the proof of Fermat’s Last Theorem in 30 hours. Anthropic’s benchmark table is drawn directly in HTML on the launch page rather than published as an image: across general coding, scientific reasoning and operating-system control, Fable 5.2 and Mythos 5.2 lead Claude Opus 5.5 and OpenAI’s GPT-6 Astra, while Mythos 5.2 posts higher scores on restricted offensive security suites including CyScenarioBench (76.3%) and CVE Rediscovery (89.6%). Independent evaluator Artificial Analysis ranked Claude Fable 5.2 first overall on its Intelligence Index v5 with a score of 65.

Key numbers

• $8 input and $40 output per million tokens (20% below 5.1), or $16 and $80 in 2.5 times fast mode
• 69.9% on FrontierSWE v2 and 49.1% on FrontierCode 2.0, ahead of 62.3% and 45.5% for Claude Opus 5.5
• 74.6% (Fable 5.2) and 75.3% (Mythos 5.2) on Terminal-Bench-Science 0.1, compared with 68.1% for GPT-6 Astra
• 89.3% on OSWorld 2.1 computer use and 64.9% on RoboDojo Sim embodied control