Anthropic releases Claude Fable 5.1 and Claude Mythos 5.1

Anthropic released Claude Fable 5.1 and Claude Mythos 5.1, point updates to the Fable 5 and Mythos 5 pair it launched in June 2026. As before, the two names refer to the same underlying model shipped with different safeguards: Fable 5.1 is generally available, while Mythos 5.1 remains restricted to organisations vetted through Anthropic’s cybersecurity and life-sciences trusted-access programmes.

The company said Fable 5.1 runs roughly 25% cheaper than Fable 5 for typical workloads and up to 45% cheaper for heavily agentic tasks, driven largely by a 75% cut to cached-input pricing. Both models support a million-token context window and outputs of up to 128,000 tokens. Anthropic also reported that cybersecurity-related interventions — cases where its safety classifiers stepped in during a session — fell by around 60% for users of Claude Code, and said Fable 5.1’s safeguards now permit the model to discover software vulnerabilities, a capability withheld from Fable 5, while still declining to help develop exploits for them.

The benchmark table Anthropic published showed the larger change. On the company’s own figures — which compared Fable 5.1 against Fable 5, Opus 5 and OpenAI’s GPT-5.6 Sol — Fable 5.1 led every axis shown, and the biggest gains were on agentic work: an agentic scientific-research test (Terminal-Bench-Science) rose to 52.6% from Fable 5’s 24.7%, and a business-workflow test (AutomationBench) to 31.4% from 17.1%, each roughly doubling. Anthropic reported 55.8% on Terminal-Bench 4.0 for agentic coding (60.9% for Mythos 5.1), an Elo of 1853 on the GDPval-AA v2 knowledge-work index against Opus 5’s 1824, 60.9% with no tools on Humanity’s Last Exam, and 41.7% under strict scoring on the newer OSWorld 2.0 computer-use benchmark. All are Anthropic’s own numbers, at updated benchmark versions, rather than independently verified results; the frontier scores table records them as reported.

Anthropic did not repeat the disputed claims from June’s launch — the autonomous genomics research and the FrontierCode benchmark result — about Mythos or Fable’s scientific output. The trusted-access restriction on Mythos, which Anthropic adopted after Mythos Preview’s limited April rollout and which survived a brief global shutdown under a Commerce Department export-control order weeks after the original launch, carries over unchanged to the 5.1 versions.