OpenAI Model Deprecation Schedule: Every Shutdown Date From December to February

The OpenAI model deprecation schedule is busy this winter. Between December 1, 2026 and February 26, 2027, OpenAI shuts down older GPT Image, GPT-5, o3, GPT-4o, text-to-speech, realtime and Whisper models. Every date below comes from OpenAI's official deprecations page, checked on November 3, 2026. Calls to a retired model will stop working, so plan your switch now.

OpenAI model deprecation schedule at a glance

Shutdown date	Models or endpoints	Suggested replacement
Nov 30, 2026	Agent Builder, Evals, v1/prompts	Agents SDK, Promptfoo, prompts in your code
Dec 1, 2026	gpt-image-1-mini, gpt-image-1.5, chatgpt-image-latest	gpt-image-2.5-sunburst or gpt-image-2.5-flare
Dec 11, 2026	gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-pro, o3, o3-pro and snapshots	Mostly gpt-5.6-sol, gpt-5.6-terra or gpt-5.6-luna
Jan 6, 2027	tts-1, tts-1-hd, gpt-4o-mini-tts	gpt-realtime-2.1-mini
Jan 14, 2027	GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini dated snapshots	gpt-6-luna
Jan 20, 2027	Legacy audio, realtime and transcription families	See the deprecations page
Feb 26, 2027	whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe, gpt-4o-transcribe-diarize, /v1/audio/translations	See the deprecations page

December 2026: image models, GPT-5 and o3

December 1 retires three image models. December 11 is the bigger change. Older GPT-5 and o3 models and their dated snapshots go, including gpt-5-pro and o3-pro. Most map to one of the three GPT-5.6 models.

January 2027: GPT-4o and older audio

• January 6: the tts-1 family goes. On the same date, active fine-tuning customers can no longer create new fine-tuning jobs.
• January 14: GPT-4o (gpt-4o-2024-08-06, gpt-4o-2024-11-20), GPT-4o mini, GPT-4.1 and GPT-4.1 mini snapshots go.
• January 20: legacy audio and realtime models go, including gpt-realtime, gpt-audio and gpt-4o-realtime.

February 2027: Whisper and transcription

February 26 retires whisper-1 and the GPT-4o transcription models. The /v1/audio/translations endpoint goes too. If you transcribe calls or podcasts, test a replacement early. Audio pipelines often hide edge cases.

How to prepare

1. Search your code for every model name in the table, including dated snapshots.
2. Check config and environment variables. Model names often live outside the code.
3. Test replacements on real prompts and compare quality, speed and cost.
4. Update fine-tuning plans before January 6.
5. Set alerts for model-not-found errors so you spot anything you missed.

What breaks if you miss a date

After a shutdown date, requests to that model name fail. Apps that hard-code a model, scheduled jobs and old integrations are the most likely to break quietly. The safest approach is to treat each date as a deadline for the switch to be live, not just started.

A simple migration test plan

• Collect 20 to 50 real prompts for each feature that uses an affected model.
• Run them on the old model and the suggested replacement.
• Compare answers, response time and token cost side by side.
• Update your prompts if the new model formats output differently.
• Ship the switch behind a flag so you can roll back quickly.

Frequently Asked Questions

What happens to fine-tuning in January 2027?

From January 6, 2027, active existing fine-tuning customers can no longer create new fine-tuning jobs, according to OpenAI's deprecations page.

When is GPT-4o being removed from the OpenAI API?

The listed GPT-4o snapshots shut down on January 14, 2027.

When is whisper-1 deprecated?

February 26, 2027.

What replaces GPT-5 and o3 in the API?

Most map to gpt-5.6-sol, gpt-5.6-terra or gpt-5.6-luna. Check the deprecations page for each model.

Summary

OpenAI's model deprecation schedule runs from image models on December 1 to Whisper on February 26, 2027. The biggest dates for most teams are December 11 for GPT-5 and o3 and January 14 for GPT-4o. Search your code now and test replacements well before each date.

Related articles

• OpenAI Agent Builder, Evals and Prompts Shut Down November 30
• ChatGPT and Codex Will Share One Usage Limit
• More AI news