Anthropic announced Claude 3.7 Sonnet on February 24, 2025, a unified frontier model that combines standard inference with optional "extended thinking" (visible multi-step reasoning) at no price premium ($3 per M input, $15 per M output, including thinking tokens). The model achieved state-of-the-art on software-engineering benchmarks (SWE-bench Verified at 63.7% baseline, 70.3% with scaffolding) and real-world agentic task frameworks (TAU-bench), with emphasis on coding and tool-use reliability rather than competition mathematics. Anthropic simultaneously launched Claude Code, a limited-preview agentic command-line tool for autonomous codebase manipulation, reporting single-pass task completion in contexts that normally require 45+ minutes of manual development time. This approach positions extended reasoning as a built-in capability choice within one model rather than a separate product tier, impacting operational cost modeling for inference workloads and enterprise vendor ROI calculations on AI-augmented software development.
364 other tracked developments touching the same claims.