Anthropic’s Claude 3.7 Sonnet: Its Smartest Hybrid AI Yet

Anthropic has unveiled its latest AI model, Claude 3.7 Sonnet, a hybrid reasoning system that represents the company’s most advanced creation to date. This new model is designed to solve complex problems and shows significant improvements in areas such as math, coding, finance, and legal tasks compared to its previous versions. Anthropic is also launching a limited research preview of an agentic coding tool called Claude Code.  This tool acts as an active collaborator, capable of reading and searching through code, editing files, running tests, committing changes, pushing code to GitHub, and even using command line tools. Claude 3.7 Sonnet is now available for users through the Claude app and for developers via Anthropic’s API, Amazon Bedrock, and Google Cloud’s Vertix AI platform.  Despite its advanced capabilities, the model is priced the same as its predecessor, costing $3 per million input tokens and $15 per million output tokens. According to Dianne Penn, Anthropic’s product research lead, reasoning is not a separate feature but an inherent quality of the AI, allowing it to handle both simple queries and more complex prompts with ease.  The model’s updated knowledge cutoff, set in October 2024, ensures it remains current, even though it does not offer real-time web search. Developers also have the option to guide the model’s thought process through its scratchpad, and they can specify the desired response time for tasks, with some responses set to as fast as 200 milliseconds, as noted by Michael Gerstenhaber, VP of product.  Internally, Anthropic employees have used Claude 3.7 Sonnet to create website designs and interactive games and to conduct lengthy coding sessions. The team even tested the model by mapping its API to control an old Pokémon game, where the new version outperformed the earlier model by defeating multiple gym leaders.