OpenAI launches GPT-5.3 Codex amid AI coding race

OpenAI has released a new AI model designed to support its recently launched coding tool, Codex.

The model, called GPT-5.3 Codex, follows the introduction of Codex earlier this week and is intended for use by software developers and other professionals. The model combines coding capabilities from earlier Codex systems with broader reasoning and professional knowledge from its GPT-5.2 model.

GPT-5.3 Codex is faster than previous versions and is built to handle longer tasks that involve research, tool use and extended execution. Users can interact with the system while it is working without losing context.

The new model was used internally during its own development, including debugging training processes, managing deployment and evaluating test results.

The model performed strongly on several internal and industry benchmarks used to measure coding and agent-based tasks, including SWE-Bench Pro and Terminal-Bench, the company said.

Notably, the release comes shortly after a competing agent-based coding model, Claude Opus 4.6, was introduced by rival Anthropic. OpenAI and Anthropic had initially planned to release their tools at the same time, though Anthropic announced its model shortly before OpenAI’s release.