MythosLM

Independent reference. Not affiliated with, endorsed by, or sponsored by Anthropic PBC. Details.

Claude Mythos Specs: Context, Output, Pricing & Retention

Specs & benchmarks informational Last reviewed Updated

The Mythos-class spec sheet

Anthropic's Mythos-class sits a tier above Opus. The two productized variants released on June 9, 2026, Fable 5 and Mythos 5, are the SAME underlying model. What differs is the safeguard configuration, not the raw capability or the core specifications below.

Confirmed specifications

  • Context window: 1,000,000 (1M) input tokens.
  • Maximum output: up to 128,000 (128k) tokens per response.
  • Standard pricing: $10 per million input tokens and $50 per million output tokens after the introductory window.
  • Covered Model: yes, Mythos-class is treated as a Covered Model, which carries additional safety and compliance handling.
  • Data retention: 30 days.
  • Zero-data-retention (ZDR): not available for Mythos-class.

A note on Preview pricing

These figures describe the generally available Fable 5 release. The earlier Mythos Preview (April 2026, via Project Glasswing, roughly 50 organizations) used different economics, around $25 input / $125 output per million tokens. Do not conflate Preview pricing with the GA rate sheet.

Availability caveat

A US export-control directive issued June 12, 2026 suspended access to both Fable 5 and Mythos 5, and that suspension is currently active. The specs remain accurate, but live access may be unavailable depending on your jurisdiction. Always confirm current status with Anthropic before planning around these numbers.

Frequently asked questions

Are Fable 5 and Mythos 5 the same model?

Yes. They share one underlying model. The difference is safeguards: Fable 5 runs classifiers that can reroute flagged requests, while Mythos 5 has those classifiers lifted and is Glasswing-only.

Does Mythos-class support zero-data-retention?

No. Mythos-class does not offer ZDR. Standard data retention is 30 days, which is an important consideration for regulated or privacy-sensitive workloads.

What is the context window and output limit?

The context window is 1M input tokens, and the model can generate up to 128k tokens of output in a single response.

Sources & further reading

Facts on this page link to their source. Quotes are kept under 15 words and attributed; figures labelled unofficial are third-party until Anthropic publishes system-card numbers.