Removed from the guide page, moved to the deprecated list
"GPT-4o" has been removed from the guide page in OpenAI's developer documentation that introduces models available through the API. On Nov. 3 (Nov. 4 Japan time), posts saying GPT-4o had been retired from the API spread on X. @aitrackerbot, a detection account that automatically tracks OpenAI's updates, pointed out that the model could still be called and had only disappeared from the documentation.
When ORIENS checked on Nov. 4, the guide page listed only "GPT-6 Astra," "GPT-6.1 Sol," "GPT-6.1 Terra," "GPT-6 Luna" and task-specific models. GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini appear in the "Deprecated models" section of the full model list, which is collapsed by default. The individual GPT-4o page remains, with its status shown as "Deprecated."
Shutdown notified in October, set for Jan. 14, 2027
OpenAI had notified developers on Oct. 14 (Oct. 15 Japan time) that it would remove GPT-4-series models from the API. The models covered are GPT-4o (the Aug. 6, 2024, and Nov. 20, 2024, versions), GPT-4o mini, GPT-4.1 and GPT-4.1 mini, with shutdown on Jan. 14, 2027. "Fine-tuned" versions that companies have customized from these models will also stop on the same day. On the official page, GPT-4o in the API was deprecated but not shut down as of Nov. 4.
The May 13, 2024, version of GPT-4o, the first one released, was shut down on Oct. 23 as announced in April. GPT-4o was removed from the chat service "ChatGPT" on Feb. 13, so the latest notice applies only to the API. "chatgpt-4o-latest," which called the version used in ChatGPT, was also shut down in the API on Feb. 17.
- GPT-4o removed from ChatGPTOfficial
- chatgpt-4o-latest shut down in the APIOfficial
- Notice of shutdown of the May 13, 2024, version of GPT-4oOfficial
- Notice of shutdown of GPT-4o, 4o mini, 4.1 and 4.1 miniOfficial
- May 13, 2024, version of GPT-4o shut downOfficial
- Detection bot points out removal from documentation guideLeak
- Scheduled shutdown of GPT-4o, 4o mini, 4.1 and 4.1 miniOfficial
Replacement Luna sharply cuts prices
OpenAI named "GPT-6 Luna" as the replacement for the four models. The company describes it as the most efficient model for high-volume workloads.
On the price list, per-million-token prices for GPT-4o are $2.50 for input and $10 for output, compared with $0.10 for input and $0.50 for output for Luna. Luna is also cheaper than GPT-4o mini, at $0.15 for input and $0.60 for output. The context window grows from GPT-4o's 128,000 tokens to 1.05 million tokens.
- GPT-4o
- GPT-4o mini
- GPT-6 Luna
Default reasoning and fine-tuning require attention
By default, Luna performs "reasoning," thinking before it answers. The default reasoning level is medium, and DataNorth reported that OpenAI's migration guide recommends starting with the no-reasoning setting. Because the number of tokens used changes with the reasoning setting, articles by DataNorth and DEV Community point out that the cost per task needs to be compared on a company's own workloads.
In the older "Chat Completions API," the mechanism for calling external functions works with Luna only when reasoning is turned off. OpenAI directs users to the "Responses API" for built-in tools such as web search.
Luna does not support fine-tuning. OpenAI is scaling back fine-tuning, and new training jobs can be submitted only until Jan. 6, 2027. Companies that use GPT-4o mini customized for their own needs have no model to move to in the same form. DataNorth listed alternatives such as writing detailed instructions, putting a few examples in the prompt and retrieving internal documents to pass to the model.
Until April 14 on Azure
The schedule differs on Microsoft's cloud "Azure OpenAI." Under Microsoft's plan, GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini will be retired on April 14, 2027. Microsoft names "GPT-5.1" as the replacement and does not list GPT-6 Luna.
Fine-tuned versions will keep running until Oct. 1, 2027, for GPT-4o and GPT-4o mini, and until Oct. 14 for GPT-4.1 and GPT-4.1 mini. The May 13, 2024, version of GPT-4o ends on Azure on Dec. 9.
In OpenAI's API, the shutdown date of each model can be checked in the "shutdown_date" field of the function that returns the list of models.






