Anthropic tells developers to retune prompts for Claude Opus 5.5
Anthropic has published prompting guidance for Claude Opus 5.5, the model it released on 22 September, and much of it asks developers to revisit settings carried over from Opus 5. The guide says Opus 5 prompts should still work without edits, but recommends testing several effort levels rather than reusing the one already in place.
Opus 5.5 runs at medium effort by default, one level below the high default on Opus 5, and thinking cannot be switched off: requests that try return an error, where Opus 5 allowed it at high effort or below. In Anthropic's testing, medium effort on Opus 5.5 matches or surpasses high effort on Opus 5 for coding and knowledge work. The guide calls effort the first setting to adjust when balancing quality against speed and cost, and suggests saving xhigh and max for tasks where they genuinely change the output.
Chat applications are advised to drop system-prompt lines telling Claude to think carefully before responding, because Opus 5.5 decides for itself how much to think. In one chat product test, removing that line made replies start sooner with no clear decline in quality. Anthropic's own documentation carries such a line for Opus 4.7, where low effort has to be kept for speed.
Other sections ask developers to re-check older workarounds, including instructions written for running with thinking off and fixes for reading charts and screenshots. An output cap set for Opus 5 with thinking off may cut responses short on Opus 5.5, since thinking consumes part of the cap even when it is not visible. For agent teams, the guide recommends a time budget based on expected task length, with the software tracking elapsed time; small groups given time signals finished research tasks faster than a solo agent without them while matching its answer quality.
Reported by Search Engine Journal ↗
Topics: Claude