OpenAI gives GPT-4o and GPT-4.1 92 days — half the notice the same family got in April
OpenAI retired two things on 14 October 2026, on opposite sides of its product line, and the pair is more informative than either one alone.
On the API, OpenAI notified developers that the remaining GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini snapshots are deprecated and will be removed on 14 January 2027. Inside the subscription products, OpenAI’s documentation states that “on October 14, 2026, GPT-5.5 will retire from ChatGPT, ChatGPT Work, and Codex on all plans, including consumer, Business, Enterprise, and Edu plans.”
One of those came with 92 days of notice. The other came with none, and did not need any — because a subscription has no deprecation policy to breach.
The API half: 92 days for a generally available model
The five IDs on the list are gpt-4o-2024-08-06, gpt-4o-2024-11-20, gpt-4o-mini-2024-07-18, gpt-4.1-2025-04-14 and gpt-4.1-mini-2025-04-14. The recommended replacement is gpt-6-luna, and OpenAI adds the line that matters most to anyone who has invested in customisation: “fine-tuned versions of these models follow the same shutdown schedule as their underlying base models.”
Now put the date next to OpenAI’s own published commitment, on the same page. Generally available models get “at least 6 months.” Specialised variants — chat, Codex and deep-research models — get “at least 3 months.” Preview models can go with as little as two weeks.
14 October to 14 January is 92 days. GPT-4o and GPT-4.1 are not previews and they are not specialised variants. They are the general-purpose chat completions models that an enormous amount of production software is pinned to.
The comparison that makes this unambiguous is inside the same model family. In April 2026, OpenAI deprecated gpt-4o-2024-05-13 with a 23 October 2026 shutdown — 184 days. Six months later, the two remaining 4o snapshots and the entire 4.1 line get half that.
Our model retirements tracker holds every dated OpenAI notice of the past year, and the pattern until today was consistent:
| Notice given | Models | Class |
|---|---|---|
| 184–179 days | gpt-4o-2024-05-13, gpt-4-turbo, o1, o3, o3-pro, gpt-5, gpt-5-pro, gpt-3.5-turbo, the audio and realtime families, dall-e | Generally available |
| 182 days | gpt-5.1, gpt-5.3-codex, gpt-5.4-nano — the 1 October batch | GA plus one Codex variant |
| 97 days | tts-1, tts-1-hd, gpt-4o-mini-tts | Audio |
| 94–87 days | gpt-5.x-chat-latest, gpt-5.x-codex, o3-deep-research, computer-use-preview, search previews | Specialised variants and previews |
| 20 days | gpt-5.4-cyber | Restricted-access variant |
Every previous sub-100-day notice went to a variant, a preview or a TTS model — exactly the classes the policy covers at three months or less. GPT-4o and GPT-4.1 are the first plain GA chat models to be given the specialised-variant clock. Two weeks ago OpenAI gave gpt-5.3-codex, a Codex variant entitled to three months, a full 182 days. The notice period is tracking something other than the model class. OpenAI’s policy does reserve the right to move faster “where safety or compliance concerns” apply; the page states no such reason here.
The replacement is a tier down, not a like-for-like
The single recommendation of gpt-6-luna for all five models flattens a real distinction. Luna is the small tier of the GPT-6 family. GPT-4o and GPT-4.1 are mid tier. At published list prices:
| Retiring model | List (in / out) | Tier | OpenAI’s replacement | Like-for-like |
|---|---|---|---|---|
| GPT-4o mini | $0.15 / $0.60 | small | GPT-6 Luna $0.10 / $0.50 | same — a 33% / 17% cut |
| GPT-4.1 mini | not separately listed | small | GPT-6 Luna $0.10 / $0.50 | same |
| GPT-4.1 | $2 / $8 | mid | GPT-6 Luna $0.10 / $0.50 | GPT-6 Sol $2 / $10 |
| GPT-4o | $2.50 / $10 | mid | GPT-6 Luna $0.10 / $0.50 | GPT-6 Sol $2 / $10 |
So the migration is a price cut for small-model work and a 25% increase in output-token cost for anyone moving GPT-4.1 work to the tier it actually belongs in. On a workload producing 50M output tokens a month, that is $400 to $500 — a $100/month increase arriving as a forced migration rather than a price rise, which is why it will not show up in anybody’s price-tracking.
Two things improve on the way across. Luna and Sol both carry 1,050,000-token context windows against GPT-4o’s 128,000, and Sol’s cached input is $0.20 against GPT-4.1’s $0.50. Both also carry the above-272,000-token surcharge that doubles input and bills output at 1.5x for the whole request — our API cost calculator applies it, and it is the one thing that can make a bigger window cost more than a smaller one.
The ChatGPT half: no notice, because no policy
On the subscription side, GPT-5.5 simply stops being selectable today, on every plan from Free to Enterprise. Paid plans are pointed at GPT-6 Sol; Free and Go plans at GPT-6 Luna.
GPT-5.5 became ChatGPT’s default model on 5 May 2026. That is a five-month life for a model that millions of people were switched onto automatically and that plenty of teams then wrote prompts, custom GPTs and Codex configurations against.
There is no policy being broken here, and that is precisely the asymmetry worth naming. OpenAI publishes a deprecation policy, a notice floor and a dated changelog for the API. For ChatGPT, ChatGPT Work and Codex there is no committed notice period at all — the retirement appears in a documentation page, and the clock is whatever OpenAI decides it is. A $200-a-month Pro subscriber has strictly weaker guarantees about model availability than a developer spending $20 on the API.
This is the second time in a month that the subscription side moved on a documentation page: ChatGPT Pro 200’s included usage halves on 30 October 2026 at the same $200 price. Prices stay still; what the price buys does not. The pricing tracker now logs both.
What this means for buyers
If you build on the API: your migration budget is the thing that got cut, not your bill. Grep for all five retiring IDs plus gpt-4o-2024-05-13, which dies on 23 October 2026 — nine days from now — and inventory every fine-tune sitting on top of them. Then benchmark Luna and Sol. The 20x input-price gap between the two makes the choice worth a day of evaluation rather than a search-and-replace.
If you pay for ChatGPT: nothing you buy changed price today, but the model picker is three names shorter and anything you pinned to GPT-5.5 re-routes now. See our ChatGPT review for the current lineup and the plan-by-plan arithmetic, and Codex if the pinning lives in your coding setup.
If you are choosing a vendor: this is a datapoint about retirement risk, which is a real cost of ownership and almost never priced. Anthropic commits to at least 60 days and sets firm dates only at deprecation; Google’s Gemini dates are explicitly “earliest possible”; OpenAI publishes the longest floor of the three and has now gone under it for a GA model. The comparison that matters is not the policy text but the dated record, which is why we keep one.
What changed in our trackers
ai-models.json:gpt-4o,gpt-4o-miniandgpt-4.1now carry deprecation 14 October 2026 and shutdown 14 January 2027 withgpt-6-lunaas the stated replacement, and a newgpt-4.1-minirow was added. Snapshot coverage is noted on each row.pricing.json: two events added — the 14 January 2027 API removal, and today’s GPT-5.5 retirement from the ChatGPT, ChatGPT Work and Codex pickers.- No price changed on either side. Anthropic’s API rate card was re-verified against the vendor page today and is unchanged.
Frequently asked questions
When exactly do GPT-4o and GPT-4.1 stop working?
The API shuts them off on 14 January 2027. The affected IDs are gpt-4o-2024-08-06, gpt-4o-2024-11-20, gpt-4o-mini-2024-07-18, gpt-4.1-2025-04-14 and gpt-4.1-mini-2025-04-14, and OpenAI states that fine-tuned versions follow the same schedule as their base models. The gpt-4o-2024-05-13 snapshot was deprecated separately in April and retires on 23 October 2026, so a codebase pinned to that one has nine days, not three months.
Is 92 days a policy violation?
It is below OpenAI's published floor for this class of model. The deprecations page commits to 'at least 6 months' for generally available models, 'at least 3 months' for specialised variants such as chat, Codex and deep-research models, and as little as two weeks for previews. GPT-4o and GPT-4.1 are generally available chat models, and 14 October to 14 January is 92 days. OpenAI's policy also allows a shorter period where safety or compliance requires it, and the page gives no reason here. Every previous sub-100-day notice in our registry went to a specialised variant, a preview or a text-to-speech model.
Is GPT-6 Luna really a drop-in replacement for GPT-4.1?
Not on capability. OpenAI names gpt-6-luna as the replacement for all five retiring models, but Luna is the small tier of the GPT-6 family at $0.10 in / $0.50 out, while GPT-4o and GPT-4.1 are mid-tier models. The like-for-like replacement for those two is GPT-6 Sol at $2 in / $10 out, which is 25% more expensive per output token than GPT-4.1's $8. So the recommended migration is a price cut only if your workload was small-model work in the first place. Re-benchmark before you re-pin.
What happened to GPT-5.5 in ChatGPT?
It retired from ChatGPT, ChatGPT Work and Codex on 14 October 2026 on every plan, including Business, Enterprise and Edu. OpenAI's documentation directs Plus, Pro, Business, Enterprise and Edu users to GPT-6 Sol and Free and Go users to GPT-6 Luna. GPT-5.5 became ChatGPT's default model on 5 May 2026, so the model a subscriber was switched onto in May was withdrawn about five months later. The API is unaffected by this particular change.
What should I do this week?
Grep your codebase for the five retiring IDs plus gpt-4o-2024-05-13, and list every fine-tune whose base model is on that list. Then benchmark against both GPT-6 Luna and GPT-6 Sol rather than accepting the single recommendation, because the two sit in different tiers and the cost difference between them is 20x on input. If you run ChatGPT or Codex rather than the API, check saved prompts, custom GPTs and Codex configurations that name GPT-5.5 — those re-route today.
Does this change what ChatGPT costs?
No. No ChatGPT plan price changed on 14 October 2026, and no API list price changed either. What changed is which models those prices buy. The nearest pending price change on the subscription side is ChatGPT Pro 200's included usage halving from 20x to 10x of the Plus allowance on 30 October 2026, at the same $200.
Sources
- OpenAI API — Deprecations (primary: '2026-10-14: GPT-4 series model deprecations' — gpt-4o-2024-08-06, gpt-4o-2024-11-20, gpt-4o-mini-2024-07-18, gpt-4.1-2025-04-14, gpt-4.1-mini-2025-04-14, shutdown 14 January 2027, replacement gpt-6-luna, 'Fine-tuned versions of these models follow the same shutdown schedule as their underlying base models'; also the stated policy of 'at least 6 months' for generally available models, 'at least 3 months' for specialised variants and as little as 2 weeks for previews)
- ChatGPT Help — Models (primary: 'On October 14, 2026, GPT-5.5 will retire from ChatGPT, ChatGPT Work, and Codex on all plans, including consumer, Business, Enterprise, and Edu plans', with GPT-6 Sol named as the replacement on Plus, Pro, Business, Enterprise and Edu and GPT-6 Luna on Free and Go)
- OpenAI API — Pricing (primary: GPT-6 Luna $0.10 in / $0.50 out, GPT-6 Sol $2 in / $10 out, GPT-4.1 $2 in / $8 out, GPT-4o $2.50 in / $10 out, GPT-4o mini $0.15 in / $0.60 out per million tokens)
Related tool reviews
Questions or corrections? Email Pick Right. Want the full list? See all news.