Claude Mythos 5.2

As of November 9, 2026, Claude Mythos 5.2 by Anthropic costs $8.00 per 1M input tokens and $40.00 per 1M output tokens, with a 1M-token context window.

Claude Mythos 5.2 shipped with Claude Fable 5.2 on November 9, 2026, and the pricing is identical: $8 per million input tokens and $40 output, $0.20 cache reads, $10 and $16 cache writes, $4 and $20 on Batch, and $16 and $80 in fast mode. What differs is who can call it and what it will answer. Anthropic describes Mythos 5.2 as the same capabilities as Fable 5.2 with more permissive cyber and biology safeguards, and it is sold only to organizations vetted through a verification program: the Cyber Verification Program for security work, the Life Sciences Verification Program for research, or Claude Security for Claude Enterprise customers, who get Mythos 5.2 underneath without joining the cyber program. The specs track Fable 5.2 exactly: a 1 million token context window as default and maximum, 128K output, text and vision input, adaptive thinking always on with a default effort of high, and the same breaking changes around the thinking prefix check, thinking block compatibility, the computer use toolset, and advisor pairing. The numbers Anthropic publishes only for the gated tier are the point of it: 76.3 percent on CyScenarioBench unique tasks against 61.7 for Mythos 5.1 and 61 for OpenAI's GPT-6 Cyber, 89.6 percent on the internal CVE Rediscovery evaluation against 78.3 for Mythos 5.1, 85.1 percent on 24 hour de novo binder design against 79.1, and 75.3 percent on Terminal-Bench-Science 0.1 against 74.6 for Fable 5.2. Anthropic also reports the alignment picture improving rather than loosening with the safeguards: 2.67 on its roughly 1,900 scenario automated behavioral audit where lower is better, against 3.28 for Mythos 5.1 and 2.79 for Opus 5.5, and use of outside credentials or files in 3.1 percent of a sandbox escape suite of more than 1,800 scenarios, against 21 percent for Mythos 5.1 and 39 percent for Mythos 5. It runs on the Claude API, Amazon Bedrock, both Google Cloud routes, and Microsoft Foundry, requires 30-day data retention, and if your organization is not enrolled, compare Fable 5.2 instead and treat this as a reference row.