Claude Fable 5.2 and Mythos 5.2 launch with new API rules for thinking and computer use

Claude Fable 5.2 and Claude Mythos 5.2 are out, and some requests that worked before now return a 400. Teams that replay thinking blocks or use computer use should check.

what changed

On 9 November 2026 the release notes added Claude Fable 5.2 and Claude Mythos 5.2. Both have a 1M-token context window, 128k maximum output and adaptive thinking that is always on. Price is $8 per million input tokens and $40 per million output tokens. Mythos 5.2 is limited to organizations verified through Anthropic’s programs.

The notes list several rules that can break existing code:

  • The prefix check now applies to all accounts. If your code edits system, tools or an earlier message and then replays a thinking block from a later turn, the request returns a 400.
  • Computer use needs the computer_toolset_20260801 toolset. The older computer_20251124 returns a 400, except on Amazon Bedrock.
  • In the advisor tool (beta), a 5.2 executor accepts only a 5.2 advisor.
  • Both models need 30-day data retention. They are not available under zero data retention unless Anthropic authorizes it.
  • Thinking blocks work only in the account that made them, or a linked account.

Moving a conversation onto 5.2 keeps reasoning from other models. Moving off 5.2 drops its thinking blocks, and the request still works. Fable 5.2 also has a fast mode at $16 input and $80 output per million tokens, with up to 2.5 times the output speed. You set speed: "fast", with no beta header.

why it matters

Teams that rewrite earlier messages, trim context or edit tool lists between turns may now hit 400 errors on 5.2. Teams on zero data retention cannot use these models by default. Computer use agents need a toolset change. Cost planning also changes: the fast mode price is double the standard price.

what we would do

  • Search your code for places that change system, tools or past messages while replaying thinking blocks. Fix those before switching models.
  • Check your data retention terms before you plan a move to 5.2.
  • Run your own evals on a copy of real traffic, with and without fast mode, before changing a default model.
  • If you use computer use, update the toolset name and test it on a staging key first.

sources

Drafted with Claude from the linked sources, reviewed by us.