On October 14, 2026, OpenAI notified developers that it will end access to some models in its GPT-4 series in the API on January 14, 2027. The affected models are GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini, and the recommended replacement for all of them is gpt-6-luna. On the same day, October 14, from 19:00 (Japan time), China's Xiaomi will also automatically redirect API requests for its MiMo-V2.5 series of AI models to the new V2.6 series. Both changes affect developers who build AI into their own apps and services.
Models OpenAI is retiring
According to OpenAI's developer Deprecations page, the following models will be removed from the API on January 14, 2027. The names in parentheses are the names used to specify the model in a program (model IDs).
- GPT-4o (gpt-4o-2024-08-06, gpt-4o-2024-11-20)
- GPT-4o mini (gpt-4o-mini-2024-07-18)
- GPT-4.1 (gpt-4.1-2025-04-14)
- GPT-4.1 mini (gpt-4.1-mini-2025-04-14)
The recommended replacement for all four is gpt-6-luna. OpenAI also says that models fine-tuned on a company's own data based on these models will become unavailable on the same schedule as the original models.
OpenAI says that when it retires a model, it notifies customers using that model by email and lists it on this page. The period from this notice to the shutdown is about three months. The same page also gives a guideline of notifying "at least 6 months" in advance for generally available models, but this time the period is shorter than that.
Note that, also in the GPT-4 family, gpt-4.1-nano and gpt-4o-2024-05-13 will be shut down earlier, on October 23, 2026, under a separate notice issued in April 2026.
What is GPT-6 Luna, the replacement?
gpt-6-luna is one of the models in OpenAI's GPT-6 series. OpenAI describes it as its "most efficient model for focused, high-volume tasks." According to the model page, pricing per million tokens is $0.1 for input ($0.01 with caching) and $0.5 for output, the amount it can handle at once (context) is 1.05 million tokens (up to 922,000 tokens of input), and a single output can be up to 128,000 tokens. However, for requests with more than 272,000 input tokens, input and cache prices double and the output price is 1.5 times higher. Writing to the cache costs $0.125 per million tokens.
gpt-6-luna is a "reasoning model" that thinks before answering, and its default thinking level is "medium." To use function calling (a mechanism for calling external functions) in the Chat Completions API, you need to set the thinking level to "none." In addition, gpt-6-luna does not support fine-tuning.
However, when the model changes, the length and style of answers and how well instructions are followed can change even with the same instructions (prompts). It is important to test with your own usage, including the pricing structure, before switching.
Impact on ChatGPT
This announcement concerns the API for developers. In the ChatGPT app, GPT-4o, GPT-4.1 and GPT-4.1 mini were already retired on February 13, 2026 (on the Business, Enterprise and Edu plans for organizations, use of GPT-4o within GPTs continued until April 3, 2026). According to OpenAI's Help Center, at that time users were told that "these models remain available in the API." Now, a shutdown date has been set for API access as well. For general users who only use ChatGPT, nothing changes in how they use it as a result of this announcement.
Xiaomi's MiMo-V2.5 is also being retired
Xiaomi has announced the retirement of its previous-generation models to coincide with the new MiMo-V2.6 series released on September 22, 2026. According to Xiaomi's API documentation page, the schedule and replacements are as follows.
- Automatic switchover: from 18:00 on October 14, 2026 (China time). The table on the page says 10:00 the same day, but the note and announcement banner say 18:00
- Retirement of old model IDs: 10:00 on October 21, 2026 (China time)
- Replacement for mimo-v2.5-pro: mimo-v2.6-pro
- Replacement for mimo-v2.5: mimo-v2.6-flash
From October 14, requests sent with old model IDs will automatically be processed by the corresponding new model, and charges will be calculated at the new model's prices. After the October 21 deadline, old model IDs will return errors. In Japan time, the automatic switchover starts at 19:00 on October 14, and the retirement takes effect at 11:00 on October 21.
There are also changes to settings. Values set in requests for mimo-v2.5 will, as a rule, carry over after the automatic switchover. However, if you have not specified the output token limit (max_completion_tokens), the default for mimo-v2.6-flash will be used, increasing from 32,768 to 131,072. Also, in thinking mode, for both V2.6 Pro and Flash, temperature and top_p, which adjust the variability of answers, cannot be set by the user and are fixed at 1.0 and 0.95 respectively. Since the thinking mode of mimo-v2.5 was already fixed at the same values, this point requires particular attention if you were setting these values with mimo-v2.5-pro.
mimo-v2.5-pro supported only text input, but the replacement mimo-v2.6-pro can also understand images and more. Xiaomi says API configuration parameters can be used as is.
What developers should do now
If a model shutdown deadline arrives without anyone noticing, an app can suddenly start returning errors one day. Checking things in the following order will give you peace of mind.
- Inventory the model IDs you use: search your program code, configuration files and environment variables for names such as gpt-4o, gpt-4.1 and mimo-v2.5. Tools built by other teams in your company and the settings of external business tools are also easy to overlook. gpt-4.1-nano and gpt-4o-2024-05-13 become unavailable earlier, on October 23, so treat them as especially urgent.
- Check shutdown dates in the model list: OpenAI includes a shutdown date (shutdown_date) field in the results of the API that returns the list of models (List models). Setting up a way to check it regularly will help you notice the next retirement early.
- Test the replacement: before switching in production, run the replacement model with the instructions and inputs you normally use, and compare whether the quality and format of the answers change.
- Check differences in pricing and behavior: if token usage, pricing or output length changes, your monthly costs will change too. Since gpt-6-luna is a reasoning model, also check the thinking level setting and the surcharge for long inputs. At the same time, check whether the configurable parameters have changed.
- Review fine-tuned models: if you use models fine-tuned with OpenAI, they will become unavailable on the same day as the original models. The replacement gpt-6-luna does not support fine-tuning, and OpenAI will also stop accepting new fine-tuning jobs from existing customers on January 6, 2027, so consider replacing them early with other approaches, such as refining instructions (prompts).
CoAI's take
GPT-4o is the model that underpinned ChatGPT's popularity and that many companies have built into their operations via the API. It remained available in the API even after it was removed from ChatGPT, but now the shutdown schedule has been made clear. With only about three months to go, including the year-end and New Year holidays, services using the GPT-4o family need to plan their migration early.
For Xiaomi's MiMo-V2.5, the automatic switchover begins on the evening of October 14, and old model IDs will stop working on October 21. Note that even during the automatic switchover, the pricing and settings of the new model apply. It has become normal for AI models to be replaced every few months. Managing model IDs in one place and preparing a procedure for testing new models in advance will become increasingly important.
