GitHub has announced that GPT-5.3-Codex, OpenAI’s latest coding model, is now rolling out on GitHub Copilot. The Microsoft-owned company said that the new model reached new high scores on benchmarks it uses during testing, with improvements in reasoning and executing in complex, long-running workflows that use tools. This has resulted in 25% faster performance over GPT-5.2-Codex on agentic coding tasks.
To use the new Codex model, you will need to be a Copilot Pro, Pro+, Business, or Enterprise customer. You can choose the model in the model picker in Visual Studio Code in all modes (chat, ask, edit, agent), github.com. GitHub Mobile (iOS and Android), GitHub CLI, and GitHub Copilot Coding Agent. If you do not see it yet, that’s because GitHub is doing a gradual rollout.
For customers who want to enable the model within organizations and use the Copilot Enterprise or Copilot Business plans, administrators will need to enable the GPT-5.3-Codex policy in Copilot settings.
The rollout of this model will help those developers most who are doing complex software architecture work and automated refactoring stand to gain the most from this model update thanks to its specialized focus on agentic workflows. As this update cuts down the time tool-driven tasks take by 25%, it helps to reduce the latency traditionally associated with large-scale AI-assisted code generation.
The deployment of this model comes after the recent general availability of Claude Opus 4.6 and the preview of Opus 4.6’s “fast mode”. Claude is very popular among developers for its coding accuracy and the latest Codex should give it some healthy competition. It’s important for GitHub Copilot to offer a variety of models too so that it can compete against other AI coding assistants.
If you try out the new model and want to give your feedback, you can head over to the GitHub Community to do so. If you rely on documentation or technical guides from GitHub, expect these to be updated following the release of GPT-5.3-Codex.