Fable 5.2 thinks at medium effort even when told not to

If you’re moving a system prompt from Claude Fable 5.1 or Opus 5.5 to Fable 5.2, Anthropic’s new prompting guide says to reread it, because some older instructions raise cost, cause side effects or stop working. A line like “answer directly without deliberating” didn’t stop Fable 5.2 from thinking at medium effort and above in Anthropic’s testing, so the guide says to remove it and lower effort instead.

The guide also offers system-prompt text for turns that end before the work is done. In sessions Anthropic replayed at medium effort, about half as many ended with doable work left, and output tokens rose by about 30%.

Anthropic suggests starting at high, the default on the Claude API, then running evals at medium and xhigh too.