Thinking Tokens

Tokens generated during a model's internal reasoning process before producing a visible response. Thinking tokens consume context budget but improve quality on complex tasks. On current Claude models, adaptive thinking steered by effort replaces fixed budget_tokens budgets, which return a 400 error on Opus 4.7 and later and on 5.x models (https://platform.claude.com/docs/en/build-with-claude/thinking).