Claude Fable 5 and Claude Mythos 5

2026-06-09 · Publisher : Anthropic · Domain : Artificial Intelligence

Anthropic launches Claude Fable 5, a Mythos-class model made safe for general use, and Claude Mythos 5, the same model with lighter safeguards, reserved for cyberdefense and biology research.

What changes

  • Capabilities superior to any previously publicly available model in software engineering, knowledge work, vision and scientific research.
  • Classifier-based safeguards (cybersecurity, biology and chemistry, distillation): the affected requests are handled by Opus 4.8; more than 95% of Fable sessions do not use it.
  • Mythos 5: same weights, access through Project Glasswing and selected biology researchers.
  • Mandatory 30-day data retention (no zero data retention). On June 12, a US export control suspended access; it was lifted on June 30 and access was restored on July 1.

Anthropic launches Claude Fable 5, a Mythos-class model made safe for general use, and Claude Mythos 5, the same model with lighter safeguards, reserved for cyberdefense and biology research. ### What changes - Capabilities superior to any previously publicly available model in software engineering, knowledge work, vision and scientific research. - Classifier-based safeguards (cybersecurity, biology and chemistry, distillation): the affected requests are handled by Opus 4.8; more than 95% of Fable sessions do not use it. - Mythos 5: same weights, access through Project Glasswing and selected biology researchers. - Mandatory 30-day data retention (no zero data retention). On June 12, a US export control suspended access; it was lifted on June 30 and access was restored on July 1. ### Specifications - Context window: 1M tokens. Maximum output: 128,000 tokens. Adaptive thinking always on. - Knowledge cutoff: not disclosed for this launch. - Availability: claude.ai, Claude Code, API, Bedrock, Google Cloud, Foundry. Identifiers: `claude-fable-5` and `claude-mythos-5`. ### Pricing | Item | Price (USD per million tokens) | | --- | --- | | Input | 10 | | Output | 50 | | Cache write 5 min / 1 h | 12.50 / 20 | | Cache read | 1 | | Batch (input / output) | 5 / 25 | ### Benchmarks No usable numeric table in the text of the announcement (charts). Results cited by the vendor: best score on FrontierCode (Cognition) and on Hebbia's financial benchmark, 3 times more access to the last act of Slay the Spire than Opus 4.8, and 9 strong candidates out of 14 targets in protein design (Mythos 5, internal evaluation). ### Key takeaways First Anthropic tier above Opus in general availability, at $10 / $50, with a more restrictive security and data retention regime.

Links

Tags : LLM, Anthropic, Claude, Fable, Mythos, Sécurité