Introducing upgrades to Codex
Summary (EN)
OpenAI announced a substantial Codex update centered on GPT-5-Codex, a version of GPT-5 optimized for agentic software engineering. The company describes the model as designed for real engineering workflows, including building projects from scratch, adding features and tests, debugging, large-scale refactors, and code review. A key product claim is dynamic behavior across task sizes: the model is meant to feel faster on small interactive requests while persisting for much longer on larger cloud tasks, sometimes working independently for hours while iterating against tests and failures. OpenAI also says GPT-5-Codex follows repository instructions such as AGENTS.md more closely and produces cleaner code without requiring long prompt scaffolding. Beyond the model itself, the announcement bundles multiple workflow changes. Codex CLI has been rebuilt around agentic coding, now supporting image inputs, progress tracking, web search, MCP connections, improved tool-call visibility, and simplified approval modes. A new IDE extension for VS Code, Cursor, and compatible editors brings the agent into local developer workflows while keeping cloud and local context connected. Codex also gets stronger GitHub integration, code review functionality intended to surface critical flaws, and infrastructure changes such as container caching and automated environment setup to reduce task latency. The release is commercially important because it turns Codex from a single coding surface into a connected product spanning terminal, IDE, cloud agent, GitHub, and mobile access. That broad integration makes OpenAI’s developer offering less about autocomplete and more about delegation of sustained engineering work.
Summary (ZH)
OpenAI 发布了面向 Codex 的一轮重大升级，核心是推出针对“智能体式软件工程”优化的 GPT-5-Codex。官方称，这一版本重点训练真实工程场景，包括从零搭建项目、补功能和测试、定位并修复 bug、执行大规模重构，以及进行代码审查等。产品层面最重要的变化之一，是模型会根据任务复杂度动态调整投入程度，在小型交互请求上更快响应，在复杂云端任务上则可以更长时间独立工作，持续迭代实现、修复测试失败并最终交付结果。OpenAI 还表示，GPT-5-Codex 对仓库内规则文件如 AGENTS.md 的遵循更好，输出代码质量更高，也减少了用户必须编写长提示词来约束风格和规范的负担。除了模型升级，这次发布同时重构了 Codex 的工作流体系。新版 Codex CLI 增加了图片输入、待办进度追踪、网页搜索、MCP 连接、更清晰的工具调用展示以及更简化的权限模式；新的 IDE 扩展把 Codex 带入 VS Code、Cursor 等编辑器，并实现本地环境与云端任务之间的上下文衔接。GitHub 集成、代码审查能力、容器缓存和自动环境搭建等基础设施改进也一并上线。这意味着 Codex 正从一个单点编码工具，转向覆盖终端、IDE、云代理、GitHub 乃至移动端的完整工程协作产品，商业意义在于它更像可持续委派工作的工程同伴，而不仅是代码补全器。
Source