Claude Fable 5.1

Claude Fable 5.1 is Anthropic’s generally available Mythos-class model for coding, knowledge work, computer use, and long-running agent tasks. It launched on September 1, 2026 alongside Claude Mythos 5.1, which uses the same underlying model with different safeguards and limited trusted access.

Anthropic describes Claude Fable 5.1 and Claude Mythos 5.1 as the same model with different safeguards. Fable 5.1 is the generally available version. Mythos 5.1 is reserved for approved cybersecurity and life-sciences users through trusted access programs.

The most important product change is not only benchmark performance. Anthropic kept the base input and output prices at $10 and $50 per million tokens while cutting cache-hit pricing from $1 to $0.25 per million tokens. The provider estimates that this makes Fable 5.1 about 25% less expensive than Fable 5 for its typical August 2026 workloads, and up to about 45% less expensive for context-heavy, tool-heavy agentic workloads.

Those percentages are provider estimates for the workloads described by Anthropic. Actual cost depends on token volume, cache reuse, effort setting, tool calls, context length, retries, and task completion.

Anthropic reports substantial improvements over Fable 5 across:

• agentic terminal coding;
• scientific research workflows;
• business and knowledge work;
• computer use;
• multidisciplinary reasoning;
• long-running autonomous tasks;
• root-cause debugging;
• code review and software maintenance.

The provider’s launch materials report 52.6% on Terminal-Bench-Science 0.1 at maximum effort and 55.8% on Terminal-Bench 4.0 for Fable 5.1. Benchmark results depend on harness, effort, safeguards, task release, tools, and evaluation setup, so use them as provider evidence rather than a universal ranking.

Fable 5.1 is designed to complete more difficult tasks while reducing the cost of repeated context reads. Compared with Fable 5, Anthropic reports higher results on its benchmark suite and better performance at lower effort settings.

The practical comparison should include:

• accepted task result;
• total tokens and cache reads;
• elapsed time;
• retries and human intervention;
• effort setting;
• test pass rate;
• safeguard interventions;
• total task cost.

A higher benchmark score does not guarantee lower project cost. The cache-price reduction matters most for long sessions that repeatedly read code, tool results, and prior context.

Fable 5.1 and Mythos 5.1 use the same underlying model. The difference is access and safeguards.

• Fable 5.1: generally available with production safeguards.
• Mythos 5.1: limited trusted access with more permissive safeguards for approved cybersecurity and life-sciences work.

Do not treat Mythos 5.1 as a public premium tier that any user can select. Anthropic controls access through its trusted programs.

Anthropic’s pricing documentation says cache hits and refreshes for Fable 5.1 and Mythos 5.1 are priced at 0.025x the base input price. Platform, region, enterprise agreement, data-residency, and cloud-provider pricing can differ, so verify the route you will actually use.

Fable 5.1 is generally available, but safeguards can still block or redirect some requests. Anthropic says its updated cybersecurity safeguards produce fewer false-positive interventions and allow defensive vulnerability discovery, while exploit development and other dual-use tasks remain restricted. Biology safeguards also remain stricter for advanced research.

Fable 5.1 can still:

• misunderstand ambiguous requirements;
• spend heavily at high effort;
• produce code that requires review;
• miss security or authorization issues;
• fail a long task after consuming substantial tokens;
• behave differently across tools and harnesses;
• require human approval for high-risk actions.

Treat long autonomous runs as monitored engineering work. Set budgets, checkpoints, tool permissions, tests, and stop conditions before execution.