Claude Fable 5.1: Longer agentic runs, cheaper cache reads, and stricter thinking-block rules

Claude Fable 5.1 and Claude Mythos 5.1

Claude Fable 5.1: Longer agentic runs, cheaper cache reads, and stricter thinking-block rules

Anthropic's Claude Fable 5.1 succeeds Claude Fable 5 at the same price, with cache reads at a quarter of the cost. It targets long-running agentic coding, multistep research, and document work, and brings a 1M-token context window, always-on adaptive thinking, and per-message effort control. Breaking changes: forced tool use is unsupported, earlier models can't read its thinking blocks, and editing earlier turns invalidates them. New betas include turn-scoped system messages and progress updates between tool calls. Text output is watermarked, and images/videos carry C2PA credentials.

To keep thinking valid across a long session, treat the conversation as append-only.

More from this day

2026-09-01