Short answer
As of November 12, 2026, you can’t switch to GPT-4o in ChatGPT: OpenAI says it was retired from ChatGPT on February 13, 2026. The API is separate. OpenAI says GPT-4o remains available through the API, but its API model page labels it deprecated. Before relying on it in an API integration, check whether it is available to your project and whether a shutdown date has been announced. (OpenAI Help Center)
If you mean ChatGPT
There is no current ChatGPT menu path that restores GPT-4o. OpenAI’s retirement notice says it is no longer available in ChatGPT. Older instructions to enable Settings → General → Show legacy models described an earlier product state; that setting does not undo the retirement. Choose a model currently offered in your ChatGPT model picker. (OpenAI Help Center)
If you mean an API integration
ChatGPT retirement does not itself change your API request. OpenAI’s API model page lists gpt-4o and marks it deprecated. The page lists gpt-4o-2024-08-06 and gpt-4o-2024-11-20 as deprecated, and gpt-4o-2024-05-13 as shut down. The page excerpt does not establish a shutdown date for the gpt-4o alias. OpenAI says deprecated models have a shutdown date and recommends checking GET /v1/models for each model’s optional shutdown_date field. Verify the current status with the same project credentials your application uses before changing a production integration. (GPT-4o model page)
Check project availability and lifecycle
Use the project’s API key to list models. OpenAI documents this endpoint as listing currently available models; its deprecation guide says the response can include shutdown_date when a date has been announced. Keep the key in an environment variable or other secret store, not in source code. (List models reference)
curl https://api.openai.com/v1/models -H "Authorization: Bearer $OPENAI_API_KEY"
Inspect the response for the gpt-4o model ID and any shutdown_date. If it is absent or unavailable to the project, do not assume that a ChatGPT subscription or setting grants API access. If a shutdown date is present, plan migration before that date.
Conditional Chat Completions smoke test
If the model appears in your project’s available list and your integration uses Chat Completions, you can make a minimal test by setting the request’s model value to gpt-4o. OpenAI lists Chat Completions as a supported endpoint for GPT-4o and documents the endpoint’s message-based request. This example uses a project API key; OpenAI recommends a project service-account identity for deployed workloads. I have not run this request. Treat it as a documented smoke test, not proof of production readiness. (GPT-4o model page)
curl https://api.openai.com/v1/chat/completions -H "Authorization: Bearer $OPENAI_API_KEY" -H "Content-Type: application/json" -d '{"model":"gpt-4o","messages":[{"role":"user","content":"Reply with the word ready."}]}'
For a successful response, inspect its top-level model field; OpenAI’s response example shows this field in the returned Chat Completion object. Compare its value with the GPT-4o identifiers on OpenAI’s model page rather than requiring an exact match to the requested alias: the page lists gpt-4o alongside dated identifiers, but does not specify which snapshot the alias selects. (Chat Completions response example · GPT-4o model page)
Before rollout, test your actual prompts, response handling, limits, and error paths. If you use another endpoint or SDK, confirm its request format and model compatibility in the API reference before changing the model setting.
If the switch fails
- GPT-4o is missing from ChatGPT: its absence is expected after the ChatGPT retirement. The old legacy-model setting is not a fix. Choose a model currently shown in ChatGPT. (OpenAI Help Center)
- GPT-4o is missing from the API model list or a request is rejected: check that you used the intended project credentials, inspect the project’s model list and any
shutdown_date, and compare the request with the API reference. Do not interpret a missing model as a ChatGPT access issue. (List models reference) - The request succeeds but your application fails: test your application’s actual prompts, response parsing, limits, and error handling in a non-production environment before rollout.
Bottom line: you can’t switch to GPT-4o in ChatGPT now. For API use, OpenAI marks GPT-4o deprecated; check project-level availability and lifecycle information before testing or relying on it in production.
Sources
Retiring GPT-4o and other ChatGPT models — OpenAI Help Center
List models — OpenAI API Reference