Why AI companies retire models at all, explained
AI companies retire old models mainly because every model they keep online uses computing capacity that newer models need. Anthropic says the cost of keeping models available "scales roughly linearly with the number of models we serve," in its Nov. 4, 2025, commitments on model deprecation. OpenAI and Google give similar reasons on their deprecation pages, and add that fewer models make choices simpler and that old ones are used less.
In one sentence
Serving a model costs about as much whether it is new or old, so companies retire old models to free capacity, cut costs and steer users to models they consider safer and more capable.
How we compared the companies
We read the official deprecation pages of Anthropic, OpenAI and Google, and the retirement announcements in which the companies explained their decisions, as of Nov. 6, 2026. Each reason below is quoted or paraphrased from those pages.
What the companies say
Anthropic: capacity. "Anthropic currently deprecates and retires models to ensure capacity for new model releases," its deprecations page says. In its commitments, the company explained that the cost and complexity of serving models grows roughly in line with how many it serves, so each extra model costs about as much as the last.
OpenAI: newer models, simpler choices, less use. "As we launch safer and more capable models, we regularly retire older models," OpenAI's deprecations page says. It also says retirements improve reliability and "make it easier for developers to choose the right models," and a Sept. 26, 2025, notice describes models "with declining usage." When it retired GPT-4o from ChatGPT, OpenAI said only 0.1% of users still chose it each day, and that the move "allows us to focus on improving the models most people use today," in its Jan. 29, 2026, notice.
Google: capacity, again. In a note about its Gemini 2.5 models on its deprecations page, Google wrote: "This helps us maintain sufficient capacity for both ongoing legacy workflows and new applications."
Why Anthropic, OpenAI and Google say they retire AI models, and the notice they promise
| Company | Main reason given | Notice promised |
|---|---|---|
| Anthropic | Capacity for new model releases | At least 60 days for publicly released models |
| OpenAI | Safer and more capable models; simpler choices; declining usage | As much as reasonably possible in safety or compliance cases |
| Capacity for legacy workflows and new applications | Advance notice; listed dates are the earliest possible |
What retiring a model costs
Anthropic is the only one of the three companies that lists the downsides on its deprecations page. Users who value specific models must migrate, researchers lose access for comparative studies, and "model retirement introduces safety- and model welfare-related risks." Its commitments describe the safety risk as one "related to shutdown-avoidant behaviors by models."
To soften those costs, Anthropic committed to keeping the weights, the trained model itself, of all its publicly released models for at least the lifetime of the company, so that it has "the ability to make past models available again in the future."
What it doesn't say
None of the three companies publishes what it costs to serve a given model, so the economics can't be checked from outside. The reasons are also general: no company explains, model by model, why a specific retirement date was chosen. And "retired" refers to the company's own service; Anthropic, for one, says it keeps the weights.
Context
How much warning users get varies. Anthropic promises "at least 60 days' notice" before retiring a publicly released model. Google says it will communicate shutdown dates "with advance notice," and that the dates it lists are the earliest possible. OpenAI says that if safety or compliance concerns require retiring a model sooner, it will give "as much notice as reasonably possible."
Retirements are also getting more frequent. On Oct. 23, 2026, OpenAI shut down more than a dozen models in a single day, from GPT-3.5 Turbo to o4-mini, according to its deprecations page.
FAQ
Why do AI companies retire old models?
Mainly to free computing capacity for new models. Anthropic says serving costs grow roughly in line with the number of models it keeps available.
Does a retired AI model get deleted?
Not necessarily. Anthropic has committed to keeping the weights of all its publicly released models for at least the lifetime of the company.
How much notice do companies give before retiring a model?
Anthropic promises at least 60 days for publicly released models. Google promises advance notice, and OpenAI says it will give as much notice as reasonably possible in safety cases.
Why did OpenAI retire GPT-4o from ChatGPT?
OpenAI said only 0.1% of users still chose GPT-4o each day and that retiring it let the company focus on the models most people use.
Sources
- Anthropic, Commitments on model deprecation and preservation (Nov. 4, 2025) · archived copy
- Anthropic, Model deprecations · archived copy
- OpenAI, API deprecations · archived copy
- Google, Gemini API deprecations · archived copy
- OpenAI, Retiring GPT-4o and older models in ChatGPT (Jan. 29, 2026) · archived copy
Cite this
Why AI companies retire models at all, explained, New Mind Museum, November 6, 2026. https://newmindmuseum.com/models/2026/11/why-ai-companies-retire-models/
New Mind Museum. (2026, November 6). Why AI companies retire models at all, explained. https://newmindmuseum.com/models/2026/11/why-ai-companies-retire-models/
“Why AI companies retire models at all, explained.” New Mind Museum, 6 Nov. 2026, https://newmindmuseum.com/models/2026/11/why-ai-companies-retire-models/.
@misc{new-mind-museum-2026-why-ai-companies-retire-models-at-all-ex,
author = {{New Mind Museum}},
title = {Why AI companies retire models at all, explained},
year = {2026},
url = {https://newmindmuseum.com/models/2026/11/why-ai-companies-retire-models/}
}Free to quote and cite with a link to this page.