Sonnet 4 Extended Thinking

Extended Thinking takes the dependable Sonnet 4 model and gives it a larger reasoning budget. Instead of replying as soon as it finds a plausible path, it decomposes the problem, checks intermediate steps, and revises its plan before producing a final answer. The result is steadier performance on multi-step analysis, trickier proofs and derivations, repository-scale coding tasks, and long-form planning, with calmer error rates and more consistent logic. You keep the familiar Sonnet 4 interface—streaming, function calls, and schema-true JSON—so it drops straight into agent pipelines and retrieval-augmented apps. It also handles text and images in the same conversation, grounding explanations in charts, documents, and screenshots when needed. The trade-off is latency and token usage: Extended Thinking is slower and chattier under the hood, which is why teams often route routine queries to standard Sonnet and reserve this mode for moments when correctness matters most.