Just now, Cursor AI announced the integration of Claude 3.7 Sonnet and a streamlined user interface for enhanced ease of use. Furthermore, Cursor has introduced cross-chat conversation functionality, automatically summarizing chat logs and carrying them over to new chat windows for an improved user experience.
Earlier, Anthropic officially released its latest reasoning model—Claude 3.7 Sonnet. This model, with its innovative design and significantly improved programming capabilities, has quickly become an industry focus. The core innovation of Claude 3.7 Sonnet lies in its fusion of rapid response and deep thinking capabilities, mimicking the human brain's thought process to provide a more fluid user interaction.
In practical application, users can choose standard mode for quick answers or switch to extended thinking mode, allowing the model to self-reflect before responding. This mode is particularly suitable for complex tasks in mathematics, physics, and programming, providing more accurate and in-depth solutions. Via API, users can precisely control the model's thinking budget, up to 128K tokens, finding the optimal balance between speed, cost, and answer quality. Unlike traditional models, Claude 3.7 Sonnet prioritizes practical business applications over pure competition performance.
In the programming field, Claude 3.7 Sonnet's performance is particularly outstanding. In a programming test, Sonnet achieved a high score of 70.3%, surpassing other well-known models such as OpenAI's o1, o3-mini, and DeepSeek R1, which scored around 49%. This result indicates Anthropic's intention to position Sonnet as a powerful coding AI, focusing on enhancing programming capabilities to meet developers' needs in handling complex codebases and full-stack updates.
Claude 3.7 Sonnet is now fully launched, supporting free, professional, team, and enterprise versions, and is available on Anthropic API, Amazon Bedrock, and Google Cloud's Vertex AI. However, free users currently cannot use the extended thinking mode. In terms of pricing, Sonnet 3.7 remains consistent with previous models: $3 per million input tokens and $15 per million output tokens, including thinking tokens.
Beyond its improved programming capabilities, Claude 3.7 Sonnet demonstrates excellent performance in other areas. For example, in the TAU-bench test, Sonnet achieved 81.2% accuracy in retail scenarios and 58.4% in aviation scenarios, significantly outperforming other models. Furthermore, Sonnet excels in instruction understanding, reasoning ability, multimodal processing, and code writing, particularly showing dramatic improvement in mathematics and science problems when extended thinking mode is enabled.