zhineng.dev
English
Get started
← All models

Claude Mythos 5.2

Published Publisher Anthropicclaude-mythos-5-2

Positioning

Claude Mythos 5.2 is Anthropic’s specialized model for cybersecurity and life-sciences research. It shares its underlying model with Claude Fable 5.2, but is available to vetted organizations working in these dual-use fields. (anthropic.com)

Capabilities and uses

  • Supports text and image input and generates text; it has a 1 million-token context window, a maximum output of 128,000 tokens, and always-on adaptive thinking. Its knowledge and training-data cutoffs are June 2026. (platform.claude.com)
  • Anthropic reports improvements in coding, reasoning, and vision. Intended applications include software engineering, vulnerability discovery and remediation, and biomedical research. On Anthropic’s internal CVE Rediscovery evaluation, it found 14.4% more vulnerabilities than Mythos 5.1. These are publisher-reported results, not guarantees of real-world performance. (anthropic.com)

Access and limitations

Access requires verification, including through Anthropic’s Cyber or Life Sciences Verification Programs. The model is listed on the Claude API and selected cloud platforms, subject to platform and eligibility conditions. (anthropic.com)

Anthropic notes that evaluations cannot reliably capture every failure before deployment. Some AI research and development capabilities were deliberately constrained during training, and a small set of related topics use a fallback model. Outputs can still be unreliable and require appropriate human review. (anthropic.com)

Pricing

Anthropic’s published Claude API prices are in USD per million tokens. Retrieved November 14, 2026:

  • Standard: input $8; output $40.
  • Prompt caching at standard rates: five-minute cache write $10, one-hour cache write $16, and cache read $0.20 per million tokens.
  • Batch API: input $4 and output $20 per million tokens, a 50% discount on base input and output rates.
  • Optional fast mode on the Claude API: input $16 and output $80 per million tokens. Fast mode cannot be combined with Batch API; caching and data-residency multipliers can apply.

Standard rates apply across the full context window. Regional data-residency options can add charges, and partner-operated cloud platforms may set independent prices. (platform.claude.com)

Release date and official sources