OpenAI announced on 6 October 2025 that Codex, its code-generation model, is now generally available after an extended private beta [source]. The model, which powers GitHub Copilot and other code-completion tools, had been accessible only through waitlisted API access since its initial release in 2021.

The general availability marks a shift in OpenAI's deployment strategy for specialised models. Codex is based on GPT-3 architecture but fine-tuned on public code repositories. OpenAI stated the model can translate natural language into code across more than a dozen programming languages, with particular strength in Python.

No pricing changes were announced alongside the availability expansion. OpenAI's announcement did not address whether the underlying model weights or training data composition have changed since the beta period. Developers who relied on Codex during beta reported variability in output quality over time, though OpenAI has not published performance benchmarks comparing beta and GA versions.

The announcement follows a pattern of OpenAI moving models from restricted to general access after prolonged testing periods. DALL-E 2 and GPT-4 followed similar trajectories. However, the company provided no timeline for when Codex might be integrated into ChatGPT or other consumer-facing products.

OpenAI noted that Codex is subject to the same usage policies as other API products, including restrictions on generating code for malicious purposes. The model remains a separate offering from GPT-4, which also has code-generation capabilities but is not optimised specifically for programming tasks.

Developers can now access Codex through OpenAI's standard API without joining a waitlist.