OpenAI Ships GPT-5.5 with Autonomous Agent Architecture, Auto-Review Mode, and Full Browser Control
GPT-5.5 is OpenAI's most aggressive bet yet on AI that does work instead of answering questions — a model designed from the ground up for tool use, self-checking, and multi-step task completion, available immediately across ChatGPT and Codex.
OpenAI released GPT-5.5 on Thursday, calling it "a new class of intelligence for real work and powering agents" — language that deliberately repositions the company's flagship product away from chat and toward autonomous task execution. The model is available now to all paid ChatGPT subscribers and in Codex, OpenAI's developer-facing coding agent.
The most technically significant feature isn't the model itself but what sits around it. As @OpenAIDevs detailed, GPT-5.5 introduces an "auto-review mode" in Codex that lets the agent work longer with fewer human approvals. A separate verification agent monitors higher-risk steps — effectively an internal oversight layer that decides when to pause and ask for permission versus when to proceed. This is the clearest production implementation yet of the "agent that checks its own work" pattern that safety researchers have been discussing for years.
Get our free daily newsletter
Get this article free — plus the lead story every day — delivered to your inbox.
Want every article and the full archive? Upgrade anytime.
No spam. Unsubscribe anytime.