Introducing GPT-5.5 | OpenAI
Summary (EN)
OpenAI announced GPT-5.5 on April 23 as its newest flagship model for agentic work in ChatGPT and Codex. The company describes the release as a step toward computer-use systems that can take on messy, multi-part tasks with less user micromanagement. According to the launch post, GPT-5.5 is designed to understand intent faster, carry more of the work itself, and persist through ambiguity while using tools, checking outputs, and continuing until a task is complete. OpenAI emphasized gains in agentic coding, knowledge work, computer use, and early scientific research. The company also said GPT-5.5 preserves GPT-5.4-level per-token latency in production while reaching higher benchmark scores and using fewer tokens on comparable Codex tasks, improving both capability and efficiency. Safety was a central part of the release. OpenAI said the model shipped with its strongest safeguard package so far, including preparedness evaluations, external red teaming, targeted testing for advanced cyber and biology capabilities, and feedback from nearly 200 trusted early-access partners. GPT-5.5 began rolling out to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex, while GPT-5.5 Pro started rolling out to Pro, Business, and Enterprise users. OpenAI said API availability would follow after additional safety and security work with customers and partners.
Summary (ZH)
OpenAI 于 4 月 23 日正式发布 GPT-5.5，并将其定位为面向 ChatGPT 与 Codex 的新一代旗舰模型，重点服务更强的 agentic 工作流。官方介绍称，该模型的目标不是单纯提升问答能力，而是让系统在面对复杂、含糊、由多个步骤组成的任务时，能够更快理解用户意图，更独立地规划执行路径，主动调用工具、检查结果，并持续推进直到任务完成。OpenAI 特别强调，GPT-5.5 在 agentic 编程、知识型工作、计算机操作以及早期科研辅助等场景中表现提升明显。性能方面，OpenAI 表示 GPT-5.5 在真实部署中保持了与 GPT-5.4 相近的单 token 延迟，同时在多个基准上获得更高分数，并且在 Codex 相关任务中通常能以更少 token 完成工作，从而兼顾能力与效率。安全方面，此次发布搭载了 OpenAI 迄今最强的一组防护措施，包括完整的 preparedness 评估、外部红队测试、针对高级网络安全与生物能力的专项测试，以及近 200 家可信早期合作伙伴的真实反馈。产品层面，GPT-5.5 已开始向 ChatGPT 和 Codex 的 Plus、Pro、Business、Enterprise 用户逐步开放，GPT-5.5 Pro 则面向 Pro、Business 和 Enterprise 用户上线，API 版本将在进一步完成安全与部署准备后推出。
Source