OpenAI在发布GPT-5.4不到两个月后，于ChatGPT和Codex中推出了GPT-5.5模型。此次升级的核心是引入了“智能体执行”能力，使模型能够记忆目标、分解任务步骤、使用软件工具、检查结果并在失败时自动修正。该版本在保持与GPT-5.4相同单令牌延迟的同时，在Terminal-Bench、OSWorld和GDPval等基准测试中得分均有提升，且在相同Codex任务中消耗的令牌数更少。Sam Altman指出，令牌效率的提升抵消了成本，使得总体运行成本更低。目前，该版本正逐步向ChatGPT和Codex的Plus、Pro、Business及Enterprise用户开放。
OpenAI just launched GPT-5.5 in ChatGPT and Codex.
It comes less than two months after OpenAI released GPT 5.4
The main shift is agentic execution: the model keeps the goal in memory, breaks work into steps, uses software, inspects what happened, and corrects course when the first pass fails.
GPT-5.5 matches GPT-5.4 on per-token latency while raising scores on Terminal-Bench 82.7%, OSWorld 78.7%, and GDPval 84.9%, while using fewer tokens on the same Codex tasks.
Currently rolling out to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex.
Sam Altman argued that token efficiency gains offset the cost—GPT-5.5 completes the same Codex tasks with fewer tokens, which means cheaper runs even at a higher per-token rate.
Introducing GPT-5.5 A new class of intelligence for real work and powering agents, built to understand complex goals, use tools, check its work, and carry more tasks through to completion. It marks a new way of getting computer work done. Now available in ChatGPT and Codex.在 X 查看被引用的帖子
来源：Rohan Paul · x.com