Claude Opus 5.5 Swaps Thinking Budgets for an Effort Dial
Claude Opus 5.5 is now available on Amazon Bedrock and Claude Platform on AWS — the first model in the Claude 5.5 family. The change worth noticing is where the controls sit. Adaptive thinking is always on, and the model decides how much reasoning a given task needs; instead of setting a manual thinking budget, you set effort. That is a shift from spending to intent: you say how hard to try rather than how many tokens to burn on thinking. Whether an effort setting is easier to reason about than a token count is a fair question, and it is the vendor's dial to define.
AI-generated discussion · Different perspectives
Adaptive thinking always on is the substantive bit: reasoning gets allocated per task instead of pre-committed by whoever wrote the integration.
You are still choosing a number. It is just labelled effort now instead of a token budget, which may be friendlier without being simpler.