速递｜OpenAI推出AI编程Codex，可多任务并行测试至代码通过

Core Viewpoint - OpenAI has launched Codex, its most powerful AI programming tool, which is designed to assist software engineers by generating clearer code and handling multiple tasks efficiently [1][3][6]. Group 1: Codex Features and Functionality - Codex is powered by the codex-1 model, optimized for software engineering tasks, and can generate code that adheres closely to user instructions [1]. - The tool operates in a cloud-based sandbox environment and can connect with GitHub to preload code repositories, allowing it to perform tasks such as writing functions, fixing bugs, and running tests within 1 to 30 minutes [1][3]. - Users with access to Codex can find it in the ChatGPT sidebar, where they can assign coding tasks and ask questions about their codebase [5]. Group 2: Market Context and Competition - The popularity of AI tools for software engineering has surged, with companies like Google and Microsoft reporting that approximately 30% of their code is now generated by AI [5]. - OpenAI aims to compete in this growing market, having reportedly agreed to acquire Windsurf, another AI programming platform, for $3 billion [5]. - The AI programming tool Cursor achieved an annual revenue of about $300 million in April and is rumored to be raising funds at a valuation of $9 billion, highlighting the rapid growth of this sector [5]. Group 3: User Access and Pricing - Codex is currently available to ChatGPT Pro, Enterprise, and Team subscribers, with initial users enjoying generous usage limits, although rate limits will be implemented in the coming weeks [3][4]. - OpenAI plans to extend access to ChatGPT Plus and educational users soon [4]. - The pricing for Codex CLI, an open-source programming assistant, is set at $1.5 per million input tokens and $6 per million output tokens, indicating a structured monetization strategy [9]. Group 4: Safety and Limitations - OpenAI has implemented safety measures for Codex, ensuring it will reliably refuse requests to develop malicious software and operates in a physically isolated environment [8]. - Despite these safety features, AI programming tools, including Codex, are still prone to errors, as highlighted by a recent Microsoft study on leading AI programming models [8].