Feb 2025
Claude 3.7 Sonnet and Claude Code — Extended Thinking and Agentic Coding
Anthropic released Claude 3.7 Sonnet, the first model to offer both a standard fast response mode and an optional Extended Thinking mode that surfaces the model's chain-of-thought reasoning to users before producing a final answer. It set a new record of 62.3% on SWE-bench Verified, the leading coding agent benchmark. Anthropic simultaneously launched Claude Code, a terminal-based agentic coding tool that could autonomously write, edit, test, and debug entire codebases.
Why it mattersExtended thinking with a configurable reasoning budget lets developers trade latency for accuracy on complex tasks without switching to a different model family.
Try itEnable extended thinking on Claude 3.7 Sonnet for your hardest reasoning prompts using the `thinking` parameter and compare accuracy vs. cost against standard mode on 20 representative examples.