OpenAI00:00Feature UpdatesOfficial Blog
OpenAI launches GPT-5.3-Codex for agents
Delegate longer coding tasks with steerable control.
Key Points
- 1Combines GPT-5 and Codex strengths
- 2Built for long-running research+execution
- 3Steerable while it works
- 4Requires strong permission boundaries
OpenAI introduced GPT-5.3-Codex, strengthening agentic coding for longer tasks that involve research and tool use. It’s designed for human steering during execution, fitting better into real workflows. Teams should pay close attention to permissions and execution boundaries when adopting it.