In this guide:
I dug through the community threads, OpenAI's status page, and the official release notes to separate what is actually known from the rumors. Here is what the error means, how to diagnose your case in 30 seconds, and the 6 fixes to try in the order that actually makes sense.
Update: October 28, 2026. OpenAI's status history still contains no incident for a Model not found error, and the only nearby incident, Widespread feature degradation on October 15, 2026, covered logins and signups in the ChatGPT mobile apps, not this error. The replacement model named in OpenAI's changelog is GPT-6.1 Sol, and the fix steps in this post are unchanged. If the error appears in an old chat, start a new chat and pick a current model from the model picker.
What the "Model not found" error actually means
Despite the scary wording, "Model not found" does not mean your account was deleted or your plan was cancelled. It means your ChatGPT session asked the server for a model it would not serve in Chat mode. Think of it like ordering a dish the kitchen no longer makes — the restaurant is open, but that item is off the menu.
The key detail from the October wave: this is a Chat-mode problem, not an account problem. Users reported that ChatGPT Work and Codex kept working on the same account, a free account (or a signed-out session) worked while the Plus account failed, and voice mode still answered while typed messages failed. When the failure follows the mode instead of the account, the fix is about the session and the model it requests — not about your login.
Why it's happening now: the October 14 GPT-5.5 retirement
On October 14, 2026, OpenAI retired GPT-5.5 from ChatGPT, ChatGPT Work, and Codex across every plan, including Business, Enterprise, and Edu. API access was not affected. The replacement is GPT-6.1 Sol (gpt-6.1-sol) — GPT-5.5 was retired from ChatGPT on October 14, and the migration has been bumpy.
The timing is hard to ignore: the first public report of the chat-side "Model not found" error appeared on X at 19:14 UTC on October 14 — just 14 minutes after the Codex retirement cutoff. Reports peaked on October 15 and grew sparse by October 16, with several users saying the error cleared on its own. Whether the wave is fully over is not known, and OpenAI has not linked the two events.
What OpenAI has actually said is one line in its ChatGPT model release notes: "We're retiring GPT-5.5 from ChatGPT, ChatGPT Work, and Codex. This change does not affect API access." That is it — nothing about the error message, nothing about old conversations, nothing about custom GPTs. On October 15, the status page did list "Widespread feature degradation" from 16:29 to 18:56 UTC, but that incident described login and signup problems in the iOS and Android apps and never mentioned "Model not found." Users kept reporting the error after that incident was marked resolved, so treat it as a separate problem.
Diagnose your case in 30 seconds
Which pattern matches you? Jump straight to the fixes listed.
- Only regular Chat fails — Work and Codex work fine on the same account. This is the account-side wave. Go to Fix 1, then Fix 5 and Fix 6.
- One browser fails while another works (for example, Chrome fails but Edge works). Your browser's stored state is broken. Go to Fix 3.
- It fails everywhere — web, desktop app, and phone. Session or server-side problem. Go to Fix 1, Fix 2, Fix 4, Fix 5.
- Conversation search fails with the same message. Same error family — it rides on the same Chat session. Start with Fix 1.
6 fixes for ChatGPT "model not found", in order
Fix 1: Start a brand-new chat and pick a current model
Old conversations and custom GPTs can keep pointing at a model that no longer exists in Chat. Earlier retirements (GPT-4o, GPT-4.5 in June 2026) moved old chats to the closest current equivalent automatically — but OpenAI has not confirmed that happened cleanly for GPT-5.5, and several users in the threads found that new chats worked while old ones failed.
- Click New chat — do not continue the failing conversation.
- Open the model picker at the top of the conversation window.
- Choose a model your plan includes: GPT-6.1 Sol, or leave it on Auto or the current default (GPT-6.1 Sol, shown as 'GPT-6').
- Send a test message.
If the new chat works, your old conversation was pinned to the retired model. Copy any context you need from the old chat and continue in the new one.
Fix 2: Try the other surface
This step is a test, not a permanent workaround — it tells you where the broken state lives.
- If the desktop app fails but chatgpt.com in your browser works, the desktop app's cached session is stale.
- If the browser fails but the mobile app works, the problem is in that browser's stored data.
- If Work mode or Codex works while regular Chat fails, the failure is server-side on Chat mode — skip to Fix 5.
Several October reporters found that switching surfaces worked around the error entirely, which is exactly why the browser-specific fixes below are worth one attempt.
Fix 3: Clear ChatGPT's site data or update the app
In Chrome (or Edge): go to Settings → Privacy and security → Site settings → View permissions and data stored across sites, search for chatgpt.com, and choose Delete data. Then sign in again. Clearing the whole browser cache also works but signs you out of everything — the site-data route is more surgical.
In the desktop app: open the Menu and choose Check for Updates, install the latest version, then fully quit and reopen the app. If the app was mid-update when the error started, an uninstall and reinstall clears the broken state.
On Android or iOS: update the ChatGPT app from the Play Store or App Store, then clear the app cache (Android: Settings → Apps → ChatGPT → Storage → Clear cache). This helped some users in the threads and did nothing for others — but it costs two minutes.
Fix 4: Sign out and sign back in
A stale or corrupted login session can keep requesting the wrong model configuration. Sign out of chatgpt.com, close every ChatGPT tab, and sign back in with a fresh session. On the desktop and mobile apps, sign out from the profile menu and sign back in. This is the classic session reset — boring, but it resolves the cases where the failure is tied to your account's session rather than OpenAI's servers.
Fix 5: Check OpenAI's status page — then wait
Open status.openai.com and look for an open ChatGPT incident. If one is listed, nothing on your device will fix the error — wait for OpenAI to resolve it and retry.
One warning: during the October wave, the status page was green for most of the reports. OpenAI never listed "Model not found" as an incident, and users kept seeing the error after the one listed incident was marked resolved. So a green status page does not prove the problem is on your side — but it does mean the only remaining option is Fix 6.
If the wave pattern matches what you see (Chat fails, everything else works, reports from other users at the same time), waiting is a legitimate strategy: several users reported the error clearing on its own by October 16 without changing anything.
Fix 6: Report it to OpenAI support with these details
If none of the above works, contact support through the OpenAI Help Center. The support bot will ask for specifics — give it these five, because vague reports get vague answers:
- The exact error text ("Model not found") and where it appears (Chat, conversation search, etc.).
- Which devices and browsers fail — and which work, if any.
- Whether ChatGPT Work or Codex works on the same account.
- Whether another account (or a signed-out session) works on the same device.
- When it started, and whether it affects old chats, new chats, or both.
That pattern list is what separates "there's a real bug on Chat mode" from "your browser is broken" — and it is exactly what OpenAI's team needs to reproduce it.
What does NOT fix it (don't waste your time)
- Resetting your password and enabling MFA. The official support bot suggested this to one user in the threads. They tried it. It did not fix the problem, and no report ties this error to account security. Skip the ritual.
- Timezone-change browser extensions. A tip circulated claiming the error comes from your browser timezone not matching your IP address. One user said changing the OS timezone fixed it; another said it did nothing. OpenAI has said nothing about it — treat it as unverified, not as a fix.
- Upgrading your plan. Free accounts and signed-out sessions worked while Plus accounts failed during the wave. This is not a subscription-tier problem.
- Reinstalling your OS or switching ISPs. When the same account fails on every network and device but works in Work mode, the problem is not your connection.
Update: October 28, 2026. GPT-5.5's retirement took effect on October 14, 2026, as OpenAI's notices said it would. The replacement OpenAI names for Codex and ChatGPT Work is GPT-6.1 Sol (`gpt-6.1-sol`); GPT-5.6 Sol remains available as a fallback, and reports of OpenAI's pricing page say GPT-6.1 Sol also became the default in the Chat tab on October 7, shown in the interface simply as "GPT-6". If you still see "Model not found", the error is not the October 15 login incident — that was a separate, resolved outage affecting logins and signups on the ChatGPT mobile apps (OpenAI status page, 16:29–18:56 UTC, "Wiidespread feature degradation"), which never mentioned this error. The fixes below still apply: pick the current default model or Auto, or GPT-6 Astra / GPT-5.6 Sol, and check the model picker first.
FAQ: ChatGPT "model not found"
Why does ChatGPT say "model not found"?
Your Chat session requested a model the server will not serve in Chat mode. Since October 14, 2026, the most likely cause is fallout from the GPT-5.5 retirement — an old conversation or session pinned to the retired model. Start a new chat and pick a current model from the picker (Fix 1 above).
Is it my account or is OpenAI down?
Check the pattern: if ChatGPT Work or Codex works on the same account, or a free account works where yours fails, the problem is on OpenAI's side of Chat mode — not your account. Check status.openai.com, but remember the October wave was never listed there.
Did the GPT-5.5 retirement break ChatGPT?
OpenAI has not confirmed a link. But the first chat-side "Model not found" reports appeared 14 minutes after the October 14 retirement cutoff, peaked the next day, and hit Chat mode specifically. The retirement is the strongest suspect; the company has said nothing beyond the one-line retirement notice.
Why does ChatGPT work on my free account but not my Plus account?
Because the failure follows the session and the model it requests, not the subscription tier. A fresh session (free account, signed-out tab) gets a clean model assignment, while your existing session may still reference the retired configuration. Signing out and back in (Fix 4) resets that state.
How long does the "model not found" error last?
During the October wave, most reports ran from the evening of October 14 through October 15, with stragglers on October 16. Some users said it cleared on its own. If the error persists for more than a day after trying the fixes above, report it to OpenAI support with the five details in Fix 6.
Still stuck? Walk through the ChatGPT not working fixes for the general outage checklist, or read what changed when GPT-5.5 was retired from ChatGPT on October 14.
Is this the GPT-5.5 retirement error?
This error wave followed OpenAI retiring GPT-5.5 on October 14, 2026. Read what the GPT-5.5 retirement changed for Codex and ChatGPT for the full timeline, and see ChatGPT not working: 8 general fixes if your problem is broader than this one error.