Anthropic says outdated prompting habits can unnecessarily slow down Claude Opus 5.5.
Updated:
Anthropic says outdated prompting habits can unnecessarily slow down Claude Opus 5.5.
Since the model already uses adaptive reasoning, the company recommends removing instructions like “think carefully” or “think deeply,” starting with Medium effort, and clearly defining the desired outcome. Anthropic's tests suggest that eliminating unnecessary reasoning instructions can speed up response times without a noticeable loss in quality.