Claude Fable 5.1 targets long-running agentic work

Anthropic introduced Claude Fable 5.1 and Claude Mythos 5.1 on September 1, 2026. They are the same underlying model with different safeguards: Fable 5.1 is generally available, while Mythos 5.1 remains limited to trusted-access programs for sensitive cybersecurity and life-sciences work.

Built for long-horizon work

Anthropic positions Fable 5.1 for demanding coding, knowledge work, multistep research, and long-running agentic tasks. The platform documentation lists a 1M-token context window, a 128K maximum output, adaptive thinking, and a high default effort.

The model is available through the Claude API and Anthropic’s listed cloud platforms. Its model page says the reliable knowledge cutoff is June 2026, so applications that need current facts should continue using retrieval or tools rather than treating the model cutoff as live knowledge.

Cache-read pricing changes the economics

Fable 5.1 keeps the listed input and output rates of Fable 5 while reducing cache reads to $0.25 per million tokens. Anthropic’s announcement says the change is expected to reduce typical token-billed workloads by about 25%, with savings potentially larger for highly agentic workloads that repeatedly reuse context.

Teams should validate those savings against their own cache-hit rate, prompt shape, tool results, and effort settings. A lower cache-read price does not automatically make a workflow cheaper if it frequently invalidates the cached prefix or expands uncached tool context.

Migration details for API users

The model documentation calls out three breaking changes for existing Fable 5 callers: forced tool use returns an error, earlier models cannot read its thinking blocks, and editing earlier turns invalidates thinking blocks. Add migration tests for tool-choice behavior, thinking-block handling, and conversation edits before changing a production model alias.

The release also adds per-message effort and turn-scoped system messages in beta, readable progress updates between tool calls, and content provenance. These controls can help orchestrators tune work during a long run, but beta behavior should be isolated behind feature flags until it is validated.

Mythos 5.1 access and safeguards

Claude Mythos 5.1 shares Fable 5.1’s capabilities and specifications but is available by invitation through Project Glasswing. Anthropic describes the two products as different safeguard tiers, not two unrelated model families. Teams should use the generally available Fable path unless they have an approved trusted-access use case.

Sources

• Anthropic — “Introducing Claude Fable 5.1 and Claude Mythos 5.1” (September 1, 2026): https://www.anthropic.com/claude-fable-and-mythos-5-1
• Claude Platform Docs — “Claude Fable 5.1”: https://platform.claude.com/docs/en/models/fable-5-1/overview
• Anthropic Help Center — “Release notes”: https://support.claude.com/en/articles/12138966-release-notes
• ClaudeDevs on X — official developer updates: https://x.com/ClaudeDevs