“全球最强编程模型”来了！Anthropic发布Claude 4，连干七小时性能稳定

Core Viewpoint - Anthropic's release of the Claude 4 series models marks a new era in AI capabilities, particularly in programming, potentially reshaping the software development industry landscape [4][17]. Group 1: Model Capabilities - Claude Opus 4 is touted as the "best programming model globally," capable of maintaining stable performance over long tasks requiring focus and effort, verified by Rakuten's 7-hour continuous operation [3][8]. - Claude Sonnet 4 shows a significant accuracy improvement, achieving 72.7% in the SWE-bench test compared to Sonnet 3.7's 62.3% [5][6]. - Both models utilize a hybrid design, allowing for immediate responses and deeper reasoning, enhancing their utility in complex coding and problem-solving scenarios [5][9]. Group 2: Extended Functionality - The new models introduce "extended thinking and tool usage," enabling Claude to utilize web searches and other tools during reasoning, improving response accuracy [11]. - Opus 4 significantly enhances memory capabilities, allowing it to create and maintain "memory files" when granted local file access, improving long-term task awareness and coherence [11][12]. Group 3: Product Launch and Integration - Claude Code has officially launched, receiving positive feedback during testing, and integrates seamlessly with platforms like GitHub Actions, VS Code, and JetBrains [12][13]. - The pricing structure remains consistent with previous models, with Opus 4 charging $15 and $75 per million tokens for input and output, respectively, and Sonnet 4 charging $3 and $15 [6]. Group 4: Competitive Landscape - The release of Claude 4 series intensifies competition among AI giants, with recent announcements from Microsoft, Google, and OpenAI highlighting the race for leading AI models [15]. - Investors are encouraged to reassess the competitive landscape, particularly Anthropic's position relative to OpenAI and Google, as the capabilities of the Claude 4 series may provide opportunities for increased market share [17].