Term
Claude Mythos 5.1
Claude Mythos 5.1 (September 2026) is Anthropics frontier model for vetted cyber and life-science organizations — the same model as Fable 5.1, only with more permissive safeguards.
Watermarking: this model marks its text
Claude Mythos 5.1 is among the models Anthropic ships with provenance marking from 2 August 2026 onwards. Two mechanisms apply: a statistical watermark in the text itself — invisibly woven into word choice, related to Google DeepMind’s SynthID-Text approach — and signed provenance metadata on generated files. Sufficiently long texts therefore carry machine-checkable evidence that Claude was involved.
Important context: the watermark yields a probability, not proof. It needs length, loses signal under heavy rewriting, and code stays largely unmarked — comments, docstrings and documentation do not. Verification runs through a detection interface that is initially available only as a private preview. Older models such as Claude Opus 5 or Sonnet 5 do not mark.
Details: AI Watermarking · Watermarks in AI Text
Claude Mythos 5.1 — explained in more detail
Anthropic released Claude Mythos 5.1 on September 1, 2026, alongside Claude Fable 5.1. Both names refer to the same underlying model — the difference lies solely in the safeguard regime: Mythos 5.1 runs under more permissive safeguards than the production version Fable and showed more aggressive capabilities in testing. The model advances the frontier in coding, knowledge work and problem-solving, with notable gains in novel mathematical and scientific reasoning, cybersecurity and biology.
Example / Practical use
Because Mythos 5.1 is highly capable in cybersecurity and biology, it is not made generally available but provided only to a small, growing set of vetted organizations through Anthropics trusted access programs — limited to verified cyber and life-science entities. In testing, the model designed experimentally validated protein binders and optimized seven open-source biological deep-learning models with inference speedups of up to 2.5x. On Terminal-Bench 4.0 Mythos 5.1 reached 60.9% versus 55.8% for Fable 5.1. Technically it shares Fable 5.1s 1M-token context window, up to 128k output tokens and always-on adaptive thinking. Pricing: 10 USD per 1M input tokens, 50 USD per 1M output tokens, cache read 0.25 USD.
Distinction
Mythos is Anthropics tier above the Opus range and follows the first Mythos preview from April 2026. The key difference from Fable 5.1 is not the model itself but access: Fable 5.1 is broadly available, while Mythos 5.1 remains limited to trusted organizations. Compared with generally accessible models such as the Sonnet or Opus range, Mythos stands for maximum capability under strict access restrictions.