Claude Fable 5.1 (Adaptive Reasoning, Low Effort, Default Fallback)

Claude Fable 5.1 is Anthropic’s flagship frontier model, released in September 2026 as the premier offering in the fifth-generation Claude lineup. Positioned as a Mythos-class model, it is situated above Claude Opus 5 and is specifically optimized for "long-horizon" agentic workflows, autonomous software engineering, and high-stakes enterprise knowledge work. The model is designed to sustain intent over multi-hour sessions, allowing it to manage complex projects such as entire codebase refactors or end-to-end scientific research with minimal human intervention.

Technically, the model supports a 1-million token context window and a maximum output capacity of 128,000 tokens. Fable 5.1 introduces a high degree of self-verification; when encountering ambiguous tasks or getting stuck, the model is designed to report its status rather than providing hallucinated success or taking shortcuts. It also represents a milestone in safety and provenance, being the first model in the Claude series to implement invisible AI watermarking on all generated text.

Capabilities of Fable 5.1 extend to advanced vision-based reasoning, specifically tuned for interpreting dense data structures within PDFs, such as financial charts, architectural diagrams, and legal tables. In coding environments, it can autonomously navigate repositories, implement high-fidelity designs, and write its own unit tests. To support highly agentic use cases, the model includes a substantial 75% reduction in prompt-caching costs compared to its predecessor, Fable 5, significantly lowering the overhead for iterative long-session tasks.