If you mean the original GPT-4, OpenAI announced that it would be fully replaced in ChatGPT by GPT-4o effective April 30, 2025. That announcement concerns ChatGPT access; it does not mean the API had the same status. If you mean GPT-4o, OpenAI retired it from ChatGPT on February 13, 2026; its temporary availability in some Custom GPTs ended April 3, 2026. A legacy-model setting cannot restore a model after retirement. (TechCrunch, reporting on OpenAI’s release notes; OpenAI Help Center)
This guidance reflects information available as of November 12, 2026. GPT-4 and GPT-4o are distinct model names; instructions for one do not necessarily apply to the other.
Can I restore GPT-4 or GPT-4o in ChatGPT?
No. OpenAI’s announcement, reported alongside its release notes, said GPT-4 would be fully replaced by GPT-4o in ChatGPT effective April 30, 2025. GPT-4o was later retired from ChatGPT on February 13, 2026. Business, Enterprise, and Edu workspaces had a temporary GPT-4o exception for Custom GPTs, but that ended April 3, 2026. (TechCrunch; OpenAI Help Center)
Older tutorials may tell you to go to Settings → General → Show legacy models. That was a historical third-party instruction for GPT-4o, not a way to restore a retired model. OpenAI says retired models no longer appear in the ChatGPT picker. For Enterprise and Edu workspaces, an admin can enable legacy access only for models that remain available; workspace settings and the picker show the current list. (OpenAI Help Center: legacy model access)
Because these models have been retired from ChatGPT, don’t treat their absence as a temporary app-display issue. A legacy-model toggle or waiting for a mobile refresh won’t reverse retirement.
What if you meant a different GPT-4-family model?
Check the exact model name in the ChatGPT picker for the account and workspace you use. If you’re in a managed workspace, ask its admin to check the workspace’s available model list and settings. A legacy-model control only exposes models that are still available; it cannot restore a retired one. OpenAI’s retirement notice specifically lists GPT-4o, GPT-4.1, and GPT-4.1 mini, so don’t assume it establishes the status of every model with “GPT-4” in its name. (OpenAI Help Center: legacy model access; model retirement notice)
Can I use the original GPT-4 or GPT-4o through the API?
API availability is separate from ChatGPT access and billing. The original GPT-4 API model page currently marks GPT-4 as shut down, so it is not a route to recover the original model. OpenAI’s GPT-4o model page lists the API identifier as gpt-4o, labels it Deprecated, and lists the Responses endpoint as supported. OpenAI’s ChatGPT retirement notice says GPT-4o continued to be available through the API; that does not mean it will remain available indefinitely. (GPT-4 model page; GPT-4o model page; OpenAI Help Center)
OpenAI’s API deprecations page says certain GPT-4-series model identifiers are scheduled for removal on January 14, 2027. The page also explains how to check a model’s shutdown_date through the Models API. Because the notice concerns specific model IDs, check the exact ID available to your API organization rather than assuming every alias and snapshot has the same shutdown date. A deprecated model is not a good default for a new production integration; plan a migration before its applicable shutdown date. (OpenAI API deprecations)
For a minimal API test with GPT-4o, create an API key, keep it on the server, and set it in your shell as OPENAI_API_KEY. The Responses API request uses the model and input fields:
curl https://api.openai.com/v1/responses \\
-H "Content-Type: application/json" \\
-H "Authorization: Bearer $OPENAI_API_KEY" \\
-d '{"model":"gpt-4o","input":"Reply with: API connection works."}'
The API uses bearer credentials; OpenAI advises keeping API keys secret and out of browser or app code. The request follows the documented Responses API pattern, but it will work only if that model is available to your API organization. If the request reports that the model is unavailable, check the exact model and shutdown status for your organization in the API documentation, then choose a currently supported model rather than changing ChatGPT settings. (OpenAI API reference; Developer quickstart)
ChatGPT subscriptions and API usage have separate billing systems. Having a ChatGPT plan does not itself establish API access or cover API usage. (OpenAI Help Center: billing)
Bottom line
- Original GPT-4 in ChatGPT: OpenAI announced it would be fully replaced by GPT-4o effective April 30, 2025; it can’t be restored there.
- GPT-4o in ChatGPT: retired in February 2026; the legacy-model toggle cannot bring it back.
- Original GPT-4 through the API: marked shut down on OpenAI’s model page.
- GPT-4o through the API: listed as deprecated. Check the exact model ID’s shutdown date and plan a migration before relying on it in production.
Sources
Retiring GPT-4o and other ChatGPT models — OpenAI Help Center
OpenAI will soon phase out GPT-4 from ChatGPT — TechCrunch
Legacy Model Access for Enterprise and Edu Users — OpenAI Help Center
API Overview — OpenAI API Reference
Developer quickstart — OpenAI API
Managing billing for ChatGPT and the API platform — OpenAI Help Center