Model retirement dates / OpenAI
gpt-4-completions
OpenAI retired gpt-4-completions on 23 October 2026. API calls to it now fail. The recommended replacement is gpt-5.6-sol.
| Provider | OpenAI |
|---|---|
| Status | Retired (as of 16 Nov 2026) |
| Shutdown | 23 October 2026 |
| Replacement | gpt-5.6-sol |
| Moving to it | API change · The old Completions endpoint (a plain prompt) doesn't serve the replacement: move the call to the Responses API or Chat Completions (messages). OpenAI's guide |
| Model type | Text |
| Announced | 22 April 2026 |
| Source | OpenAI deprecations page |
Before you switch
A replacement model answers the same prompt differently: tone, length, format, and whether JSON still parses. Driftline Pro runs your real prompts on gpt-5.6-sol in CI, so you see the new answers before you ship the switch. See a drift test, or the verdict in a fix pull request.
Check your code for it
Run this in your repository. No install and no API key; it lists every OpenAI, Anthropic, Google, Mistral, xAI, Cohere, DeepSeek, Groq, Bedrock, Vertex AI and Azure model ID it finds, with its retirement date and replacement:
npx -y @aidriftline/driftline .
Or look for this ID only: grep -rn "gpt-4-completions" . (misses aliases, Bedrock and Vertex names, and IDs split across lines).
To be warned automatically before the next retirement, add the free Driftline GitHub Action. It annotates the exact line in pull requests and can fail the build on retired models.
Where a new name is the whole change, Driftline can make it for you: npx -y @aidriftline/driftline fix --dry-run shows the edits, and the Action's fix-pr: true opens one pull request with them, plus a checklist for the lines that need more. See an example pull request.
Same date at OpenAI
ft-o4-mini-2025-04-16 gpt-3.5-turbo gpt-3.5-turbo-0125 gpt-3.5-turbo-completions gpt-4 gpt-4-0613 gpt-4-0613-completions gpt-4-1106-preview gpt-4-turbo gpt-4-turbo-2024-04-09 gpt-4-turbo-completions gpt-4.1-nano
Checked against OpenAI's official page on 16 November 2026. Something wrong? hello@aidriftline.com