OpenAI API deprecations: shutdown dates, October 2026 to January 2027

OpenAI API deprecations now come in waves, and the next one lands on October 23, 2026. That day OpenAI shuts down legacy GPT-3.5, GPT-4, GPT-4 Turbo, o1, o3-mini and o4-mini snapshots, plus gpt-image-1. On November 30 the v1/prompts API, reusable prompt objects, Agent Builder and the Evals platform go. December retires older image models and the first GPT-5 and o3 snapshots. January ends text-to-speech and the legacy audio and realtime models. After each shutdown date, requests to those models fail. If your product calls any of them, plan the switch now.

What changed

OpenAI published these notices between April 22 and October 1, 2026 on its deprecations page. A deprecated model still answers today, but it has a fixed shutdown date. OpenAI names a replacement for each one. The OpenAI model shutdown dates in this alert run from October 23, 2026 to April 1, 2027.

• Legacy GPT snapshots, October 23, 2026: gpt-3.5-turbo, gpt-4, gpt-4-1106-preview, gpt-4-turbo, gpt-4.1-nano, gpt-4o-2024-05-13, o1, o1-pro, o3-mini, o4-mini and gpt-image-1, with their dated snapshots and completions variants. Fine-tuned models on GPT-3.5 Turbo, GPT-4, GPT-4.1 nano, o4-mini, babbage-002 and davinci-002 go the same day. Replacements are gpt-5.6-sol, gpt-5.6-terra or gpt-5.6-luna, and gpt-image-2.5-sunburst or gpt-image-2.5-flare for images.
• Reusable prompts, Agent Builder and Evals, November 30, 2026: the v1/prompts API and reusable prompt objects shut down. Agent Builder shuts down, and ChatKit stays. The Evals dashboard and API shut down, and existing evals turn read-only on October 31.
• Older image models, December 1, 2026: gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest. The replacement is gpt-image-2.5-sunburst or gpt-image-2.5-flare.
• First GPT-5 and o3 snapshots, December 11, 2026: the GPT-5 snapshot deprecation covers gpt-5-2025-08-07, gpt-5-mini-2025-08-07, gpt-5-nano-2025-08-07 and gpt-5-pro-2025-10-06, and the same notice covers o3-2025-04-16 and o3-pro-2025-06-10. Their replacements are gpt-5.6-sol, gpt-5.6-terra and gpt-5.6-luna; the pro models are replaced by gpt-5.6-sol with reasoning.mode: pro.
• Text-to-speech and new fine-tuning jobs, January 6, 2027: tts-1, tts-1-hd and both gpt-4o-mini-tts snapshots shut down, with gpt-realtime-2.1-mini as the replacement. The same day, existing customers can no longer create new fine-tuning jobs. OpenAI says inference on existing fine-tuned models stops only when the underlying base model is deprecated.
• Legacy audio, realtime and transcription, January 20, 2027: gpt-realtime, gpt-4o-realtime, gpt-realtime-mini and gpt-4o-mini-realtime are replaced by gpt-realtime-2.1 or its mini version. gpt-audio, gpt-4o-audio, gpt-audio-mini and gpt-4o-mini-audio are replaced by gpt-audio-1.5. The gpt-4o-mini-transcribe-2025-03-20 snapshot is replaced by gpt-4o-mini-transcribe-2025-12-15.

OpenAI's stated minimum notice is six months for generally available models and three months for specialized variants. Preview models can be retired at much shorter notice, such as two weeks.

Who is affected

• Any product, internal tool or script that sends one of the model IDs above. Look in environment variables, config files, prompt libraries, background jobs and code copied from older tutorials.
• Teams that keep prompts as reusable prompt objects in the OpenAI dashboard and call them by ID. Those IDs start with pmpt_.
• Teams that built agents in Agent Builder, or that run their tests on the Evals platform.
• Voice features: read-aloud, phone and voice agents, live captions. The text-to-speech shutdown and the OpenAI realtime model deprecation both land in January 2027.
• Teams with fine-tuned models on the listed base models, and teams that plan more fine-tuning runs.
• Anyone who calls an alias such as gpt-5. OpenAI's model pages mark which aliases and snapshots are deprecated, so check the page for each model you use.

What to do now

1. Make an inventory. Search your code, config and logs for every model ID you send, and for pmpt_. Write down which feature uses which model.
2. Put each model ID in one config value. One setting per feature means the next migration is a one-line change, not a code hunt.
3. Test the replacement on your own examples. Keep a small set of real prompts and expected answers in your repo. Check output format, tool calls, refusals, latency and cost before you switch.
4. Move prompt objects into code. The OpenAI prompts API deprecation ends in a shutdown on November 30. OpenAI's guide, Migrate from prompt objects, shows the change. You also gain normal code review and version history for prompts.
5. Move off Agent Builder and Evals. OpenAI points to Migrate from Agent Builder for the Agents SDK route, and to a cookbook on moving evals to Promptfoo. Existing evals turn read-only on October 31, so export what you need before the November 30 shutdown.
6. Plan fine-tuning early. If you rely on fine-tuning, schedule your last training runs before January 6, 2027.
7. Watch for failures by model. Alert on failed API calls grouped by model name, so a missed call shows up in minutes, not in support tickets.

Key dates

1. October 23, 2026legacy GPT-3.5, GPT-4, GPT-4 Turbo, GPT-4.1 nano, o1, o3-mini and o4-mini snapshots, gpt-4o-2024-05-13, gpt-image-1 and the related fine-tuned models shut down.
2. October 31, 2026existing evals become read-only.
3. November 30, 2026v1/prompts, reusable prompt objects, Agent Builder and the Evals dashboard and API shut down.
4. December 1, 2026gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest shut down.
5. December 11, 2026the first GPT-5, GPT-5 mini, GPT-5 nano, GPT-5 pro, o3 and o3-pro snapshots shut down.
6. January 6, 2027tts-1, tts-1-hd and gpt-4o-mini-tts snapshots shut down; no new fine-tuning jobs for existing customers.
7. January 20, 2027legacy audio, realtime and transcription models shut down.
8. February 26, 2027whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe and gpt-4o-transcribe-diarize shut down.
9. April 1, 2027gpt-5.1, gpt-5.3-codex and gpt-5.4-nano shut down.

How we help with OpenAI API deprecations

Changing a model ID takes minutes. Knowing the new model still gives your users the right answers takes testing. Through our AI integration services, we map every model call in your product, build a small test set from your own examples and run the switch with you. If a missed shutdown has already broken a live feature, bring in an emergency software developer through our crisis response service. Dates for other providers, runtimes and databases sit in our software end-of-life tracker for startup stacks.