Core Insights - The article discusses the significant upgrade of Claude AI, introducing two new models: Claude Opus 4 and Claude Sonnet 4, which set new standards in coding, advanced reasoning, and AI agent capabilities [2][3]. Model Performance - Claude Opus 4 and Sonnet 4 have achieved impressive scores in various benchmarks, with Opus 4 scoring 72.5% in Agentic coding and 79.4% in SWE-bench, while Sonnet 4 scored 72.7% and 80.2% respectively [5][19]. - Opus 4 can independently run for 7 hours, showcasing its strong performance in long-duration tasks, while Sonnet 4 is optimized for reasoning and efficiency [6][14]. User Experience - Users have reported high satisfaction with Claude 4, particularly praising Opus 4 for its ability to generate high-quality content without manual intervention [6][7]. - The models are designed to handle complex tasks, with Opus 4 demonstrating significant improvements in maintaining context and understanding user instructions [6][15]. New Features - Anthropic has introduced new functionalities in its API, including code execution tools, MCP connectors, and file APIs, enhancing the capabilities of developers using Claude 4 [12][23]. - The pricing for the models remains unchanged, with Opus 4 priced at $15 per million tokens for input and $75 for output, while Sonnet 4 is $3 for input and $5 for output [12]. Strategic Focus - Anthropic has shifted its focus from chatbots to enhancing Claude's ability to execute complex tasks, such as programming and research, which has led to significant advancements in the models [46][48]. - The company has secured a $2.5 billion credit facility to support its AI research and development efforts, reflecting confidence in its strategic direction [56].

图片来源： Unsplash OpenAI 于 4月 14 日， 推出了名为 GPT-4.1 的新模型系列，新 AI 模型专注于编码。 GPT-4.1 面世之际，正值 OpenAI 的竞争对手如谷歌和 Anthropic 加紧研发高级编程模型。 谷歌最新发布的 Gemini 2.5 Pro 同样具备 100 万 token 上下文窗口，在主流编程基准测试中名列前 茅。 Anthropic 的 Claude 3.7 Sonnet 与中国 AI 初创公司深度求索升级的 V3 模型亦表现不俗。 包括 OpenAI 在内的许多科技巨头的目标，是训练能够执行复杂软件工程任务的人工智能编码模型。 正如首席财务官 Sarah Friar 上个月在伦敦的一次科技峰会上所言， OpenAI 的宏伟抱负是创造一 个"自主软件工程师"。该公司宣称，其未来的模型将能够端到端地编程整个应用程序，处理质量保 证、错误测试和文档编写等方面。 GPT-4.1 是朝着这个方向迈出的一步。 "我们根据直接反馈优化了 GPT-4.1 ，以在实际应用中改进开发者最关心的领域：前端编码、减少无 关的编辑、可靠地遵循格式、遵守响应结构和顺序、一致的工具 ...