Anthropic has released Claude 3.7 Sonnet, its first "hybrid reasoning model" that can solve more complex tasks and outperforms previous models in areas such as math and coding.
In addition to the new model, Anthropic is also introducing a “limited research version” of its “agent” coding tool called Claude Code. The tool is positioned as an “active collaborator” capable of searching and reading code, editing files, writing and running tests, committing and publishing code to GitHub, and using command-line tools.
Claude 3.7 Sonnet is now available in-app and for developers via the Anthropic API, Amazon Bedrock, and Vertix AI on Google Cloud. The model costs the same as the previous 3.5 Sonnet — $3 per million input tokens and $15 per million output tokens.
While OpenAI and other companies offer separate so-called reasoning models, Anthropic’s head of product research, Diana Penn, notes that the company has sought to simplify the experience of using the model. Anthropic believes that reasoning is a feature of AI, not something separate.
Penn notes that Claude 3.7 Sonnet is much better at “agent coding,” as well as solving financial and legal problems. Additionally, Anthropic gives developers the ability to control the model’s reasoning via a notebook and set time limits on responses.