Claude 3.7 Sonnet

Claude 3.7 Sonnet was the first "hybrid reasoning" frontier model: standard and extended-thinking modes selectable per request, billed at the same input/output rates but with extended thinking consuming additional output tokens for the visible reasoning trace. The release in February 2025 paired the model with Claude Code, Anthropic's official agent harness, which made 3.7 Sonnet the default backend for serious coding-agent work through 2025. The SWE-Bench Verified score (70.3% at release) was the headline number: a closed reasoning model that could be asked to fix a real bug in a real repository and succeed at a rate that materially exceeded prior frontier models. The agentic story matters because it established "the model that can act in your codebase" as a distinct product category from "the model that can answer questions about your codebase", and Anthropic was the first frontier lab to commercialize that distinction. Open-weights catch-up arrived with Kimi K2 and DeepSeek's later releases, but Claude 3.7 Sonnet held the agentic-coding leadership position long enough to define what the category looked like.