Anthropic has unveiled its latest AI model, Claude 3.7 Sonnet, which represents a significant advancement in “hybrid reasoning.” This model is designed to tackle more complex challenges and has shown improvements in areas such as mathematics and coding compared to earlier versions. In conjunction with the new model, Anthropic is also testing a coding tool named Claude Code. This tool is described as an “active collaborator” that can perform tasks like searching and reading code, editing files, writing and executing tests, and managing code on platforms like GitHub.
Starting Monday, Claude 3.7 Sonnet is available through the Claude app and can be accessed by developers via Anthropic’s API, Amazon Bedrock, and Google Cloud’s Vertex AI. The operational costs remain consistent with the previous model, set at $3 per million input tokens and $15 per million output tokens. Unlike other companies that provide separate reasoning models, Anthropic aims to offer a more integrated experience. Dianne Penn, the product research lead, explained that the company views reasoning as a fundamental aspect of AI rather than a distinct feature. This means that Claude can quickly respond to simple queries while also managing more complex tasks, like planning a detailed trip.
Claude 3.7 Sonnet has shown notable improvements in coding, finance, and legal tasks. Although it does not yet support real-time web searches, its knowledge is updated to October 2024. Developers can influence how the model operates by using its scratchpad and setting response time limits. Internally, employees have utilized the model for various projects, including website designs and interactive games. Notably, Claude 3.7 has even demonstrated its capabilities by successfully navigating challenges in a Pokémon game, outperforming its predecessor. As the AI landscape evolves rapidly, Claude 3.7 Sonnet positions Anthropic at the forefront of this competitive field.
