Original "Codex" API Shut Down with Just Three Days' Notice, Citing GPT-3.5/GPT-4 Integration
On March 23, 2023, OpenAI discontinued API access to its original Codex-family models (code-davinci-002, code-davinci-001, code-cushman-002, and code-cushman-001), which had been offered as a free beta since 2021. The company cited the fact that its successor general-purpose models, GPT-3.5 and GPT-4, now outperformed Codex even at code generation tasks — but the astonishingly short three-day transition window between developer notice and shutdown drew criticism that it had “shaken confidence in OpenAI as a reliable platform.”
Details
- Discontinued models: Four models — code-davinci-002, code-davinci-001, code-cushman-002, and code-cushman-001 — were removed from the API
- Recommended migration path: OpenAI recommended migrating to GPT-3.5-Turbo or GPT-4, judging that general-purpose models capable of handling not just code but also documentation, error messages, and software design context outperformed a code-only model
- Short notice: Developers were given only about three days between notification and the actual shutdown, sparking strong backlash from those who had integrated the Codex API into existing products
- A casualty of its free-beta status: Codex had been offered as a free beta ever since its 2021 release, never having become a formal paid product — a status likely contributing to how readily it was discontinued on short notice
- Partial revival later: After the full shutdown, access was later partially restored for some researchers through the “OpenAI Researcher Access Program,” in response to demand for research use
What happened next
The abrupt end of the original Codex made clear that OpenAI was prioritizing investment in general-purpose large models over single-purpose specialized ones. The “Codex” brand disappeared entirely at this point — only to make a dramatic comeback roughly two years later, in 2025, as an entirely redesigned cloud-based coding agent called “OpenAI Codex.”