Claude 3.7 Sonnet

Claude 3.7 Sonnet (API name claude-3-7-sonnet-20250219) launched on 24 February 2025 and was, by Anthropic's description, the first hybrid reasoning model on the market. That claim is the reason this profile exists. Every other vendor at the time shipped reasoning as a separate product line — a fast chat model on one side, a slow deliberating model on the other, with the user choosing between two different endpoints. Anthropic put both into one set of weights: standard mode answered near-instantly, extended thinking mode reflected step by step before replying, and API callers could cap the thinking budget at any number of tokens up to the 128,000-token output limit, trading cost and latency for answer quality on a dial rather than a switch. The design argument was explicitly about how people work: a single brain handles both the quick reply and the long deliberation, so a model should too. The commercial consequence was that reasoning cost nothing extra — $3 per million input tokens and $15 per million output, the same as Claude 3.5 Sonnet, with thinking tokens billed as ordinary output rather than at a premium. Anthropic also said it had deliberately optimised less for maths and computer-science competition problems and more for the tasks businesses actually run, which showed in the benchmark mix: strong real-world coding and agentic tool use rather than record contest scores. The same announcement introduced Claude Code as a limited research preview — the terminal tool that went on to become a product line of its own. Anyone tracing where agentic coding assistants came from lands on this date. Anthropic notified developers on 28 October 2025 that the model would be retired, and switched it off on 19 February 2026, almost exactly one year after launch. The recommended replacement is claude-sonnet-4-6.

▸Videos

No videos yet.

▸Reviews