OpenAI announced on 21 April 2026 that it is expanding access to Codex, its code-generation model, to enterprise customers worldwide [source]. The company stated that Codex will be available through its API with new pricing tiers and support for additional programming languages.

According to the blog post, Codex has been updated to handle "more complex codebases" and now supports over 30 programming languages, including Rust, Go, and Kotlin. OpenAI said the model has been trained on a larger dataset of public repositories and internal code samples provided by pilot customers.

The announcement follows reports from developers in recent months that Codex outputs had become less consistent, with some users observing increased hallucination of non-existent functions and libraries. OpenAI did not address these reports in the blog post, nor did it specify whether the enterprise version differs from the public API offering.

Enterprise customers will receive dedicated support channels and the option to fine-tune Codex on proprietary codebases, according to the post. OpenAI stated that all fine-tuning data will remain isolated and will not be used to train future versions of the model.

The company did not disclose which enterprises have signed up for the service or provide benchmarks comparing the updated Codex to previous versions. Pricing details were described as "available upon request" for organisations with more than 500 developers.

OpenAI said the rollout will begin in May 2026, starting with customers in North America and Europe.