Claude 3.7 Sonnet is a multimodal large language model that introduces a hybrid reasoning architecture. This design allows the model to function both as a standard language model for near-instant responses and as a specialized reasoning engine capable of extended, step-by-step "thinking." It is the first model in the Claude family to offer users and developers fine-grained control over inference-time compute by allowing them to set a specific token budget for internal reasoning during a single session.
Capabilities and Performance
The model demonstrates significant advancements in complex tasks such as software engineering, advanced mathematics, and logical deduction. In its extended thinking mode, Claude 3.7 Sonnet self-reflects and plans before providing an answer, which enhances its performance on benchmarks like SWE-bench and AIME. The internal chain-of-thought is made visible to the user, providing transparency into the model's logic and decision-making process.
In addition to its reasoning capabilities, the model maintains high performance in multimodal understanding and instruction following. It features a 200,000-token context window and is optimized for agentic workflows, where it can autonomously navigate multi-step processes and use external tools with high reliability.