Claude Fable 5.1 (and Claude Mythos 5.1)

What it is

Claude Fable 5.1 is a generally available model from Anthropic, released on 1 September 2026 with the API ID claude-fable-5-1. Claude Mythos 5.1 is the same model with different safeguards, and it is limited to trusted-access programmes. ^[1][3][5]

Fable 5.1 has a 1M-token context window, a maximum output of 128K tokens, and adaptive thinking that is always on. Its knowledge cutoff is June 2026. ^[2]

Background

Anthropic's benchmark tables compare Fable 5.1 with Claude Fable 5. ^[1]

Main features

Price. Fable 5.1 is priced at $10 per million input tokens and $50 per million output tokens, the same as Fable 5. Cache reads are 75% cheaper, at $0.25 per million tokens. Anthropic estimated savings of about 25% on typical workloads, and up to about 45% on highly agentic work. ^[1][4]

Benchmarks. Compared with Fable 5, Fable 5.1 scored 55.8% against 42.0% on Terminal-Bench 4.0, and 52.6% against 24.7% on Terminal-Bench-Science. On GDPval-AA v2 it reached 1853 Elo against 1723, and on CursorBench 73.4% against 70.5%. On Humanity's Last Exam it scored 60.9% without tools and 65.0% with tools. ^[1]

Mythos 5.1. Mythos 5.1 scored 60.9% on Terminal-Bench 4.0. It is offered through the Cyber Verification Program and the new Life Sciences Verification Program, initially only to organisations in the United States, and it powers Claude Security. ^[1]

Science. Anthropic gave two examples. Mythos 5.1 designed protein binders with a hit rate of nearly 50% across 12 targets, where a typical rate is 10 to 15%. Fable 5.1 built an elevation map of Venus covering a third of the planet at 2 to 3 km detail, which was released under a Creative Commons licence. ^[1]

Cyber and biology safeguards. Fable 5.1 can identify software vulnerabilities but cannot develop exploits. Cyber false positives fell by about 60%, and benign biology queries trigger safeguards 85% less often. ^[1]

Provenance marking. Anthropic said models released after 2 August 2026 carry an invisible text watermark, under the EU AI Act transparency code of practice that the company signed in July 2026. ^[1]

Defaults and availability. The default effort level is High in Claude Code and Medium in Cowork and Claude.ai. The model is available on Amazon Web Services, Google Cloud and Microsoft Azure. ^[1]

Position in the timeline

Anthropic's documentation recommends the model for demanding reasoning and long-horizon agentic work. Its retirement is not scheduled before 1 September 2027. ^[2][3]