OpenAI API retires GPT-4, GPT-4 Turbo, o1 and o3-mini

OpenAI shuts down a large batch of older models in its API on October 23, 2026. The list includes GPT-4, GPT-4 Turbo, GPT-3.5 Turbo, o1, o1-pro, o3-mini, o4-mini and the image model gpt-image-1. Any app that still calls one of them by name stops working once its shutdown takes effect.

OpenAI announced the batch on April 22, 2026, according to its deprecations page. The page says that at the time of shutdown, "the model or endpoint will no longer be accessible."

Which models shut down and what replaces them

OpenAI names a replacement for each retired model. Most move to gpt-5.6-sol. Layer3Labs and BenchLM, two independent trackers, confirm the main mappings.

Retired model (and aliases)	OpenAI's suggested replacement
gpt-4-0613 (gpt-4)	gpt-5.6-sol
gpt-4-turbo-2024-04-09 (gpt-4-turbo)	gpt-5.6-sol
gpt-4o-2024-05-13	gpt-5.6-sol
o1-2024-12-17 (o1)	gpt-5.6-sol
o1-pro-2025-03-19 (o1-pro)	gpt-5.6-sol with pro reasoning mode
o3-mini-2025-01-31 (o3-mini)	gpt-5.6-sol
gpt-3.5-turbo-0125 (gpt-3.5-turbo)	gpt-5.6-terra
gpt-image-1	gpt-image-2.5-sunburst or gpt-image-2.5-flare

o4-mini and gpt-4.1-nano are in the same batch. Check OpenAI's page for their replacements before you switch.

BenchLM counts 29 first-party model IDs dated October 23. That includes six fine-tuned model IDs. Fine-tuned versions follow the schedule of the model they were built on, OpenAI says.

What happens to calls after shutdown

OpenAI's page says only that a retired model "will no longer be accessible." It does not name an error code, BenchLM notes.

Layer3Labs reports that requests to a retired model ID get an HTTP 400 or 404 response with an invalid-model error. Either way, the result is the same: a failed request where your app expected an answer.

Microsoft's Azure runs on its own schedule. BenchLM reports that on Azure, o1 and o3-mini retire on November 19, 2026, and gpt-4o-2024-05-13 on December 9. On Azure, Microsoft suggests gpt-5.6-terra for o3-mini rather than gpt-5.6-sol. A retired Azure deployment returns "410 Gone," according to BenchLM's summary of Microsoft's rules.

How much warning OpenAI gives

OpenAI's page sets minimum notice periods. Generally available models get 6 months. Chat, Codex and deep-research variants get 3 months. Preview models can get about 2 weeks.

You do not have to watch the page by hand. OpenAI says the models endpoint carries an optional shutdown_date field on each model. A script can read it and warn you before a model you use goes away.

The next shutdowns

October 23 is the first of several dates. OpenAI's page lists these next:

Date	What shuts down	Replacement
November 30, 2026	The v1/prompts API and Agent Builder	Prompts in app code; Agents SDK
December 11, 2026	gpt-5, gpt-5-mini, gpt-5-nano, o3 and o3-pro snapshots	gpt-5.6-sol, terra or luna
January 6, 2027	tts-1 and tts-1-hd	gpt-realtime-2.1-mini
January 14, 2027	gpt-4o, gpt-4o-mini and gpt-4.1 snapshots	gpt-6-luna
February 26, 2027	whisper-1 and gpt-4o-transcribe	gpt-transcribe or gpt-live-transcribe

The January 14 shutdown was announced on October 14, 2026. Layer3Labs notes it covers the API only, not ChatGPT.

What this means for developers

If you have not moved off these models, treat it as an outage, not a deprecation warning.

• Search your code and config for the old names. Look for gpt-4, gpt-4-turbo, gpt-3.5-turbo, o1, o3-mini and o4-mini, including in environment variables and prompt files.
• Test the replacement before you swap it in. gpt-5.6-sol is a different model. Run your evaluation set on it, since outputs, costs and speed can all change.
• Check fine-tuned models. A fine-tune built on a retired base model goes away with it. Plan to retrain on a supported base.
• Add a shutdown check to CI. Read the shutdown_date field for every model ID your code uses and fail the build when one is close.
• Mark January 14, 2027 now. The gpt-4o and gpt-4.1 snapshots are the next big batch. Start testing gpt-6-luna before then.
• Azure users have more time. If you call these models through Azure, your dates are November 19 and December 9. Do not plan around OpenAI's date by mistake.