OpenAI GPT-5.3 leaks online, launching soon

OpenAI continues to develop its language models. A recent pull request in the company’s public Codex repository reveals references to a new model variant labeled gpt-5.3.

The pull request implements longest prefix matching for remote model slugs. Developers update the system to prioritize the most specific match when resolving model names. A new integration test ensures that requests for “gpt-5.3-codex-test” correctly map to “gpt-5.3-codex” rather than the base “gpt-5.3”.

This change indicates active internal preparation for models bearing the 5.3 designation, likely an incremental upgrade with specialized capabilities, such as enhanced coding performance.

AI insiders on X (formerly Twitter) highlight the discovery as evidence of an imminent launch. They anticipate improvements in areas like mathematical reasoning and research acceleration. Predictions for the release timeline vary, with some expecting an announcement within days.

On the other hand, the development fits OpenAI’s established pattern of regular model iterations. Previous versions, including GPT-5 and its variants, introduced significant advances in reasoning, speed, and reliability.

CEO Sam Altman has signaled ongoing progress in public appearances. In a recent CNBC interview, he hinted at the near-term release of new models, reinforcing expectations for continued rapid advancement.

OpenAI has not issued an official statement on these observations. The company typically announces major updates through its channels when ready.