News
Fable 5.1 and Mythos 5.1 Quietly Broke `tool_choice` and Cut Cache Reads to a Quarter
The September 1, 2026 release notes for Claude Fable 5.1 and Mythos 5.1 confirm both keep the $10/$50 per million token price and add a 1M-token default context with 128k max output, while dropping cache read cost to $0.25 per million, a quarter of the rate on every other Claude model. The breaking change buried in the same notes: `tool_choice` types `any` and `tool` now return HTTP 400, so callers must move to strict tool use or structured outputs. Thinking blocks are also only preserved for equal or newer models, and replaying them to an earlier model gets the block rejected or dropped.
↳ Follow the thread