Anthropic said in a guide that Claude Opus 5.5 is now the standard model for Pro, Max, Ultra and Team subscribers, and that old prompt habits no longer help. Thinking is always on. The model decides how much to think from the request and the effort setting, so a "think hard" line adds nothing. Anthropic said dropping such lines made replies start sooner, with no clear quality loss. The Decoder said these phrases date from early reasoning models.
Effort sets how long the model thinks. Anthropic said higher settings cost more without always giving better work. On its own coding and knowledge-work tests, it said, Opus 5.5 at Medium, the default, matched or beat Opus 5 at High. The guide also quotes Factory, an early tester, reporting a similar result with 20 to 25% fewer output tokens. Anthropic picked that quote. No independent study is reported.
For buyers, a model upgrade changes what a good prompt is, so effort defaults now affect cost. Anthropic said higher-priced models use plan limits faster, and Sonnet 5.5 at Extra high or Max comes close to Opus 5.5 in cost per task. Teams can remove "think hard" lines from stored prompts and compare cost per finished task at Medium with their current setting.