OpenAI API Deprecation Calendar
OpenAI retires two things: model snapshots (dated versions stop serving, replaced by a newer snapshot) and whole API surfaces (Assistants, Videos, Reusable Prompts, Evals, Agent Builder, fine-tuning). The surface shutdowns are the ones that break code you cannot fix by bumping a model string - they need a rewrite, so start those early. Every row below links to OpenAI's official deprecations page; confirm your exact snapshot there before migrating.
2026
| Date | What retires | Do this before the date | Src |
|---|---|---|---|
| Jul 23 | Shutdown wave 1 of the 2026-04-22 announcement - 18 IDs, and they are current-generation, not legacy: gpt-5-chat-latest, gpt-5-codex, gpt-5.1-chat-latest, gpt-5.1-codex, gpt-5.1-codex-max, gpt-5.1-codex-mini, gpt-5.2-codex, computer-use-preview (+ -2025-03-11), o3-deep-research (+ -2025-06-26), o4-mini-deep-research (+ -2025-06-26), gpt-4o-search-preview-2025-03-11, gpt-4o-mini-search-preview-2025-03-11, gpt-4o-mini-tts-2025-03-20, gpt-audio-mini-2025-10-06, gpt-realtime-mini-2025-10-06 | Do not skim this one because the word "legacy" is attached to it. If you run Codex-family coding agents, computer use, deep research, or the search-preview models, you are in wave 1. Repoint: codex/chat-latest to gpt-5.5, computer-use and the mini tier to gpt-5.4-mini, deep-research to gpt-5.5-pro, gpt-audio-mini/gpt-realtime-mini/gpt-4o-mini-tts to their 2025-12-15 snapshots or gpt-audio-1.5. The gpt-3.5-turbo/gpt-4/o1/o3-mini tier is wave 2 (Oct 23), not this one. | ✓ |
| Aug 10 | gpt-5.2-chat-latest, gpt-5.3-chat-latest | Move to gpt-5.5. If you pinned a -chat-latest alias for stability, note it always tracks the newest snapshot - pin a dated snapshot instead if you need reproducibility. | ✓ |
| Aug 26 | Assistants API - full shutdown (threads, runs, assistants objects) | Rewrite onto the Responses API + Conversations API. Budget real time: the object model differs (no server-side threads/runs loop), so this is a port, not a config change. Export any assistant definitions and file attachments now. | ✓ |
| Sep 24 | Videos API + sora-2/sora-2-pro - shutdown, no first-party successor | There is no drop-in OpenAI replacement. Evaluate Google Veo 3.1 or Gemini Omni Flash video now and abstract your video calls behind an interface so the swap is one adapter, not a rewrite. | ✓ |
| Sep 28 | Shutdown wave 2 of the 2025-09-26 announcement (legacy GPT snapshots) | Same drill as Jul 23: repoint pinned snapshots to current GPT-5.x. Check the deprecations page for the exact snapshot list in this wave. | ✓ |
| Oct 23 | Shutdown wave 2 of the 2026-04-22 announcement - 24 IDs, this is the genuinely old tier: gpt-3.5-turbo, gpt-3.5-turbo-0125, gpt-3.5-turbo-completions, gpt-4, gpt-4-0613 (+ completions variants), gpt-4-1106-preview, gpt-4-turbo, gpt-4-turbo-2024-04-09, gpt-4.1-nano (+ dated), gpt-4o-2024-05-13, o1, o1-2024-12-17, o1-pro (+ dated), o3-mini (+ dated), o4-mini (+ dated + ft-o4-mini-2025-04-16), gpt-image-1; the gpt-image-1 preview on the Azure mirror retires the same day | This is the wave that kills the old workhorses, including fine-tuned o4-mini - tuned weights die with the base, so retrain or replace before the date. Repoint to gpt-5.5, gpt-5.4-mini/nano, gpt-5.5-pro, gpt-image-2. | ✓ |
| Oct 31 | Evals platform goes read-only (full shutdown Nov 30) | Export eval definitions and historical run data before it locks. Move your eval harness to code (Agents SDK evals or your own runner) so grading is version-controlled, not platform-bound. | ✓ |
| Nov 30 | Shutdown: v1/prompts (Reusable Prompts), Evals platform, Agent Builder | Move Reusable Prompts into your own app code or a prompt file in the repo - stop depending on server-hosted prompt IDs. Rebuild Agent Builder flows on the Agents SDK. | ✓ |
| Dec 1 | gpt-image-1-mini, gpt-image-1.5, chatgpt-image-latest | Repoint image generation to gpt-image-2. Re-test prompts: image models shift style and parameter handling between versions. | ✓ |
| Dec 11 | Snapshots: gpt-5/gpt-5-mini/gpt-5-nano (2025-08-07), gpt-5-pro (2025-10-06), o3, o3-pro | These are the first-generation GPT-5 snapshots. Move to gpt-5.5, gpt-5.4-mini/nano, gpt-5.5-pro. Re-run your regression evals - behavior and pricing both shift across the jump. | ✓ |
2027 (announced so far)
| Date | What retires | Do this before the date | Src |
|---|---|---|---|
| Jan 6 | Fine-tuning: no new jobs accepted (final stage of the 2026-05-07 wind-down) | If your product depends on OpenAI fine-tuning, this is the hard stop for creating new tunes. Migrate the workload now: prompt engineering + few-shot, distillation to a model you control, or fine-tuning on another provider. Existing deployed tunes keep serving until their own base retires - check the model list above. | ✓ |
Not a shutdown, but it changes your bill
PRICE: GPT-5.6 makes explicit cache writes cost 1.25x the input rate. If you write to the prompt cache aggressively, your cache economics changed - re-check whether caching still nets a saving at your hit rate before assuming it does.
Waves and dates: OpenAI groups some shutdowns into waves (Jul 23 / Oct 23, and the 2025-09-26 announcement split across Sep 28). The wave label does not track how old a model is - the 2026-04-22 announcement puts current-generation Codex and chat-latest aliases in the first wave (Jul 23) and the gpt-3.5/gpt-4/o1 tier in the second (Oct 23). Always verify your exact model ID against the official deprecations page rather than assuming by age. This calendar is maintenance guidance, not legal advice.
Correction (2026-07-20): until today this page and our Week 29 digest described the Jul 23 wave as the gpt-3.5-turbo/gpt-4/o1/o3-mini tier. That was wrong - those retire Oct 23. The Jul 23 wave is the current-generation list above. Corrected against the source on 2026-07-20; details in the Week 30 digest.