Anthropic Releases Claude Fable 5.1 and Mythos 5.1

Anthropic has released Claude Fable 5.1 and Claude Mythos 5.1, marking an update three months after the Fable 5 line was introduced in June 2026. These two models share the same underlying architecture, differentiated by distinct safeguard layers. Claude Fable 5.1 is now generally available under the identifier claude-fable-5-1, while Claude Mythos 5.1 remains accessible only to vetted organizations. Both models boast a substantial 1 million token context window and a maximum output of 128,000 tokens, with adaptive thinking capabilities enabled by default. A key performance metric highlighted by Anthropic is the model's score of 52.6% on Terminal-Bench-Science 0.1. This represents a significant improvement over previous versions, with Fable 5 scoring 24.7% and Opus 5 achieving 29.0%. For comparative context, GPT-5.6 Sol scored 22.4% on the same benchmark. Anthropic notes a standard error of 3.5 to 4.5 points across models, advising users to consider the margin of error when interpreting rankings. On Terminal-Bench 4.0, Fable 5.1 achieved a score of 55.8%, and Mythos 5.1 reached 60.9%. The company candidly disclosed that the performance difference between Fable 5.1 and Mythos 5.1, despite being the same core model, is attributable to the cost of safeguard interventions, a departure from typical industry transparency. Beyond scientific benchmarks, Fable 5.1 also achieved a score of 73.4% on CursorBench 3.2.0. On Humanity’s Last Exam, it scored 60.9% without tools and 65.0% with tools. AutomationBench yielded a score of 31.4%, OSWorld 2.0 recorded 41.7% in strict mode, and GDPval-AA v2 achieved 1853. A notable commercial improvement is a 75% reduction in cache read costs, decreasing from $1.00 to $0.25 per million tokens. Anthropic estimates this translates to approximately a 25% lower cost for typical workloads and up to a 45% reduction for agentic tasks. The base input and output pricing remains unchanged at $10 and $50 per million tokens, respectively. The cost reduction is specifically tied to the cache reads, which are now priced at 0.025 times the base input cost, compared to 0.1 on other Claude models. Claude Fable 5.1 is accessible through various platforms including the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud, and Microsoft Foundry. Claude Mythos 5.1 is restricted to vetted US organizations participating in Project Glasswing.