Just a note that ChatGPT 5.5 just ran for 13.4 hours flat — invoking Claude and Gemini when I’d asked, to review Codex’s own work — without stopping.
That’s an absurd run to sustain with an out-of-the-box LLM. No coding orchestration from my side. No harness of mine keeping it awake, no watcher nudging it along. It had the job and the instruction to get its work reviewed by two other labs, and it did that, for 13.4 hours.
The model is three days old. OpenAI released GPT-5.5 on Thursday, calling it its “smartest and most intuitive to use model” yet and rolling it out to Plus, Pro, Business and Enterprise users in ChatGPT — seven weeks after GPT-5.4. Greg Brockman, OpenAI’s president, told Fortune the new capabilities “feel like they’re setting the foundation for how we’re going to do computer work going forward, or how agent computing at scale will work,” and called the model “a faster, sharper thinker for fewer tokens compared to something like 5.4.”
Thirteen hours is what that sentence looks like on a desk. The work itself was Codex’s — OpenAI’s coding agent, which “runs locally on your computer” and signs in on a ChatGPT plan — and the reviewers were the other two labs, called in on request.
A model that will call two competitors to check its own work, because it was told to, and keep going after they answer, is doing something closer to running a shop than answering a prompt.
No coding orchestration from my side. That’s the note.
