Claude 3.7 Sonnet

Claude 3.7 Sonnet marks Anthropic's entry into the reasoning model space with a hybrid approach that sets it apart from its predecessors. The model introduces a toggleable Thinking Mode that reveals its chain-of-thought reasoning process, allowing users to switch between rapid responses and extended step-by-step analysis depending on task complexity. This dual-mode design reflects a broader industry shift toward unified user experiences in chat-based large language models. The architecture particularly shines in coding scenarios, with Anthropic highlighting gains in front-end development and full-stack application updates. Evaluations show the model excels in agentic workflows, demonstrating the ability to autonomously navigate multi-step processes and complete tasks that would require significant human time.

METR's autonomy evaluation revealed particularly striking capabilities in AI research and development tasks, where Claude 3.7 Sonnet reached performance comparable to median human experts despite receiving substantially less time per task attempt. In broader autonomous capability assessments, the model demonstrated a 50% success rate on tasks that typically require about an hour of expert attention. The extended reasoning mode delivers enhanced accuracy in mathematics, coding, and instruction-following tasks while maintaining standard mode performance parity with earlier versions. Benchmark data from the model's official evaluations places it at rank 5 on instruction-following benchmarks and rank 14 on competition mathematics problems, underscoring its effectiveness across both precise instruction execution and complex analytical challenges.