Anthropic's New Model Can Answer Instantly or Stop to Think, in One Program

Claude 3.7 Sonnet combines a normal chatbot and a step-by-step reasoning model into a single system that decides — or is told — how long to think. It leads the field on real-world coding tasks.

Anthropic released Claude 3.7 Sonnet today, alongside its new Claude Code tool, and the model itself has a design worth noting. Until now, the reasoning programs that think before answering — OpenAI's o1 and o3, DeepSeek's R1 — have been separate from the fast chatbots that answer immediately. Claude 3.7 is both. It responds quickly to ordinary questions and, when a problem is hard or when you ask it to, spends time reasoning through it first. You can even dial how long it thinks.

The company says this reflects how it thinks the technology should go: not a menu of different models for different jobs, but one system that allocates effort the way a person does, thinking harder about harder things. On tests of real-world software engineering — fixing actual bugs in actual open-source projects, rather than solving competition puzzles — it leads all comers by a clear margin, which is the benchmark that has come to matter most as coding becomes the technology's proving ground.

For the three-way race among OpenAI, Google, and Anthropic, it is another lead change, and another sign that the improvements are coming as much from cleverer design as from raw size. For everyone else, the practical upshot is a chatbot that no longer makes you choose between fast and thoughtful.

Anthropic has quietly become the lab that many programmers reach for first. A model that is both quick and, when needed, careful — released the same day as a tool that puts it to work — is how it intends to stay there.