Claude's reasoning capabilities have evolved significantly. Opus 4.6 and Sonnet 4.6 introduced Extended Thinking with explicit budget_tokens. However, Opus 4.8 (May 2026) continues the paradigm shift — replacing explicit thinking budgets with Adaptive Thinking.
thinking: {"type": "enabled", "budget_tokens": N} returns a 400 error on Opus 4.7 and Opus 4.8. You MUST use thinking: {"type": "adaptive"} instead. Opus 4.8 also uses adaptive thinking and outperforms the old explicit budgets on all benchmarks.On pre-4.7 models, you set budget_tokens (minimum 1024). These tokens are consumed from your max_tokens limit. If you set max_tokens: 4096 and budget_tokens: 2048, the model has exactly 2048 tokens left for its response.
With adaptive thinking, the model dynamically decides how much to reason based on task complexity. Simple questions get instant answers; complex coding tasks trigger deep multi-step reasoning. You control intensity via the effort parameter instead of raw token counts. Fast Mode provides up to 6x speed at higher rates for latency-critical applications.
Claude Fable 5 (June 2026) takes adaptive thinking further — it is always on with no configuration required. The model dynamically allocates reasoning depth based on task complexity, achieving frontier-level performance on complex coding, scientific reasoning, and multi-step agent workflows.
Claude can perform interleaved thinking — reasoning in between sequential tool calls. This allows the model to analyze tool outputs, adjust its strategy, and deliberate before making the next action. Critical for complex multi-step agent workflows.
In Opus 4.8, thinking content is hidden by default in API responses. You must explicitly set thinking: {"type": "adaptive", "visible": true} to see the reasoning chain. This change improves response cleanliness for production applications.