I was talking to ChatGPT normally in a conversation, and it randomly switched from 5.4 thinking to 4o mini in a message without any notice. Now I can't get back to 5.4 thinking, or any other latest model like 5.3 instant through toggling the settings. New chats also use 4o mini. What should I do?
#Random model switching
pastel kettle
hazy saffronBOT
Hi @pastel kettle. I'm kapa.ai, your friendly support bot. I will try to answer your question by searching through the knowledge base. :wave:
It sounds like something is wrong with your model picker or workspace settings—what you describe (being forced onto GPT‑4o mini and unable to select GPT‑5.3 / 5.4) is not expected behavior based on the current model lineup.
From the available documentation:
- GPT‑4o and related models have been retired from ChatGPT as of February 13, 2026, except as legacy options in specific business/enterprise/edu contexts. [Retiring models; Legacy access]
- The default experience for logged‑in users should be GPT‑5.3 Instant, with GPT‑5.4 Thinking/Pro available via the model picker on eligible paid plans. [GPT‑5.3/5.4]
The knowledge sources do not explain a scenario where a normal user is suddenly locked to 4o mini with no way back to 5.3/5.4, so this likely reflects an account‑specific bug, a workspace configuration issue, or a temporary guardrail/limit that isn’t documented here.
Given that, the best next steps are:
- Check the model picker and “Configure” menu
- Make sure you’re not in a “legacy models only” or similar configuration.
- On eligible paid plans, you should see options like Instant, Thinking, and possibly Pro, and can toggle automatic switching and legacy models there. [Model picker; Enterprise model picker]
-
If you’re on an Enterprise/Edu/Business workspace
- Ask your workspace admin to check Workspace settings → Models and any role‑based access control (RBAC) settings; admins can enable/disable access to GPT‑5.3 Instant and GPT‑5.4 Thinking/Pro, and misconfiguration can block you from using them. [Enterprise models]
-
Contact OpenAI Support with details
Since the docs don’t cover this exact behavior, you’ll likely need direct help:- Go to the Help Center and use AI Chat or AI Phone Call (1‑888‑GPT‑0090 in supported regions). [Retiring models]
- Include:
- Your plan type (Free, Plus, Pro, Business, Enterprise, Edu)
- Screenshots of the model picker and settings
- A link or timestamp for the conversation where it switched from 5.4 Thinking to 4o mini
They can check whether your account hit an internal guardrail, a workspace policy, or a bug, and restore access to GPT‑5.3 / 5.4 if appropriate.
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
pastel kettle
@hazy saffron I am on a personal plus plan, and my settings are definitely set to latest, 4o mini is not even an available option. Hvae I hit some sort of usage quota for 5.4 thinking?
hazy saffronBOT
Let me check the knowledge base for you again. One moment! :books:
You might be running into limits, but what you’re seeing (being forced to 4o mini and unable to re‑select GPT‑5.4 Thinking or 5.3 Instant) is not described in the docs for Plus, so I can’t confirm it from the knowledge sources alone.
What the docs do say:
- On Plus, you can use GPT‑5.3 Instant with a cap of 160 messages every 3 hours; after that, chats “switch to the mini version of the model until the limit resets.” [Usage limits]
- For GPT‑5.4 Thinking, Plus users can manually select it with a limit of up to 3,000 messages per week. Once you hit that, “GPT‑5.4 Thinking will no longer be selectable from the menu.” Automatic routing from Instant to Thinking can still happen and does not count toward that weekly limit. [Usage limits]
The articles do not mention 4o mini as a fallback for Plus users in the current GPT‑5.3/5.4 era, and 4o/4o‑mini are listed as retired from ChatGPT (except for specific legacy contexts). [Retiring models]
Given that:
- It’s plausible you’ve:
- Hit the 5.4 Thinking weekly limit, which would explain why you can’t re‑select it, and
- Hit the 5.3 Instant 3‑hour limit, which normally triggers a switch to a “mini” model.
- However, the docs don’t explicitly say that this “mini” is 4o mini, nor that Plus users should see 4o mini at all now—so the exact behavior you’re seeing is not documented.
What you can do now
- Wait at least 3 hours and see if GPT‑5.3 Instant becomes selectable again; if you’ve only hit the Instant cap, it should reset on that schedule. [Usage limits]
- If you’ve been using Thinking heavily, you may have hit the 3,000‑per‑week cap; in that case, it won’t reappear until your weekly window resets. [Usage limits]
- Because the forced switch to 4o mini and its persistence across new chats is not explained in the docs, I recommend contacting OpenAI Support:
- Use the Help Center’s AI Chat or AI Phone Call (1‑888‑GPT‑0090) and include:
- That you’re on Plus
- That your model picker is set to latest, 4o mini is not listed, but chats are using it
- Approximate time and link to the conversation where the switch happened [Retiring models]
- Use the Help Center’s AI Chat or AI Phone Call (1‑888‑GPT‑0090) and include:
From the available information, I can’t definitively say you’ve hit a specific quota that should cause exactly this 4o‑mini behavior, so support will need to check your account directly.
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
pastel kettle
@hazy saffron I have definitely not hit the 5.4 thinking weekly limit, since the option is still selectable in my menu
hazy saffronBOT
Let me check the knowledge base for you again. One moment! :books:
You’re right to rule out the 5.4 Thinking weekly limit: the docs say that once you hit that cap, GPT‑5.4 Thinking “will no longer be selectable from the menu”, so if it’s still selectable, you have not exhausted that quota. [Usage limits]
From the knowledge sources:
- Plus users can:
- Use GPT‑5.3 Instant with a 160‑messages‑per‑3‑hours cap, after which chats “switch to the mini version of the model until the limit resets.” [Usage limits]
- Manually select GPT‑5.4 Thinking up to 3,000 messages/week. [Usage limits]
- GPT‑4o and related models are documented as retired from ChatGPT, except for specific legacy/enterprise contexts. [Retiring models]
The docs do not describe:
- 4o mini being used as the “mini” fallback for Plus users in the current GPT‑5.3/5.4 setup, or
- A situation where your settings show GPT‑5.4 Thinking selected but the conversation actually runs on 4o mini.
So, based on the available information, I cannot explain this behavior from documented limits alone; it looks like either:
- An undocumented fallback/routing behavior, or
- A bug or account‑specific issue.
Given that:
- You haven’t hit the 5.4 Thinking weekly limit (it’s still selectable).
- The 4o mini behavior isn’t documented for Plus in the current model lineup.
The next concrete step is to contact OpenAI Support so they can inspect your account:
- Go to the Help Center and use AI Chat or AI Phone Call (1‑888‑GPT‑0090). [Retiring models]
- Include:
- That you’re on ChatGPT Plus.
- Screenshots showing:
- The model picker with GPT‑5.4 Thinking selected.
- The message header or system info showing responses coming from 4o mini.
- Approximate time and a link to the conversation where the switch occurred.
From the documentation alone, I can’t give a definitive cause or self‑service fix; support will need to verify whether this is a routing bug, an internal safeguard, or something misconfigured on your account.
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
pastel kettle
@hazy saffron thank you for helping, it works now that I set 5.4 thinking to extended, weird
hazy saffronBOT
Let me check the knowledge base for you again. One moment! :books:
Glad it’s working again.
The docs do mention that when you select GPT‑5.4 Thinking, you can choose different “thinking time” levels (Light/Standard/Extended/Heavy depending on your plan), and that this setting affects how long and how deeply the model reasons before responding. [GPT‑5.3/5.4]
However, the knowledge sources don’t explain a case where switching from one thinking‑time level (e.g., Standard) to another (Extended) would fix an issue with the wrong model being used, so I can’t say why that change resolved it. It may have effectively “reset” the routing for that chat, but that behavior isn’t documented.
If it happens again, capturing a screenshot and the time it occurs and sharing that with Support would give them something concrete to investigate.
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
pastel kettle
@hazy saffron the issue is, it only works on extended thinking, not on regular thinking
hazy saffronBOT
Let me check the knowledge base for you again. One moment! :books:
The knowledge sources don’t describe any behavior where GPT‑5.4 Thinking only works correctly on Extended but not on Standard/regular, so I can’t explain this with documented settings.
What is documented:
- When you select GPT‑5.4 Thinking, you can choose a thinking‑time level in the composer:
- Plus/Business: Standard (default) and Extended
- Pro: also Light and Heavy
Your choice is saved for future queries until you change it. [Thinking time toggle]
- These levels are meant to trade off speed vs. depth of reasoning, not to change which base model is used. The docs don’t say that Standard vs Extended should route you to different model families. [Thinking time toggle]
Since:
- You’re on Plus,
- 5.4 Thinking is selectable,
- And behavior differs between Standard and Extended in a way that isn’t documented,
this likely reflects a bug or routing issue, not an intended feature.
Concrete next steps:
- Capture evidence while it’s happening
- Take screenshots showing:
- Model picker set to Thinking → GPT‑5.4 Thinking.
- Thinking‑time set to Standard vs Extended.
- The model label above responses (e.g., showing 4o mini vs 5.4 Thinking).
- Take screenshots showing:
- Contact OpenAI Support
- Use the Help Center’s AI Chat or AI Phone Call (1‑888‑GPT‑0090) and attach the screenshots plus timestamps and chat links. [Retiring GPT‑4o]
- Explain that:
- On Standard thinking time, chats appear to run on the wrong/legacy model.
- On Extended, they correctly use GPT‑5.4 Thinking.
From the available documentation, there’s no user‑side setting beyond what you’ve already tried that should fix this; support will need to inspect your account and routing configuration directly.
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
pastel kettle
@hazy saffron so I should definitely contact support right now?
hazy saffronBOT
Let me check the knowledge base for you again. One moment! :books:
Yes.
The behavior you’re seeing (GPT‑5.4 Thinking only working as expected on Extended but not on Standard, and sometimes routing to 4o mini) is not described anywhere in the docs, so it’s likely a bug or account‑specific routing issue rather than something you can fix yourself. [GPT‑5.3/5.4]
To contact support:
- Go to help.openai.com and use the chat bubble (AI Chat) or the phone option (1‑888‑GPT‑0090 where available). [How to contact support]
- Include:
- That you’re on ChatGPT Plus.
- Screenshots showing:
- Model picker set to GPT‑5.4 Thinking.
- Thinking time set to Standard vs Extended.
- The model label showing 4o mini vs 5.4 Thinking.
- Approximate timestamps and links to affected chats.
That’s the best way to get this investigated and properly fixed.
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
pastel kettle
@hazy saffron what is the quota for extended thinking mode? is it the same as the general 5.4 thinking quota?
hazy saffronBOT
Let me check the knowledge base for you again. One moment! :books:
The knowledge sources don’t mention any separate quota for Extended (or Light/Standard/Heavy) thinking time. All thinking‑time levels are treated as modes of GPT‑5.4 Thinking, not as different models with their own limits.
For Plus and Business, the documented limit is:
- Up to 3,000 messages per week with GPT‑5.4 Thinking, selected manually from the model picker. Once you hit this, “GPT‑5.4 Thinking will no longer be selectable from the menu.” [Usage limits]
For thinking time:
- When you select GPT‑5.4 Thinking, you can choose Standard and Extended (Plus/Business), and Pro users also get Light and Heavy. These are described only as speed vs. depth trade‑offs, not as separate quota buckets. [Thinking time toggle]
So, based on the docs, Extended uses the same overall GPT‑5.4 Thinking quota; there’s no documented additional or different quota specifically for Extended.
Note: you can ask me a follow up question by @ mentioning me again :speech_balloon:
pastel kettle
@hazy saffron the AI support bot keeps insisting it's not 4o mini and doesn't help, are there other support options?
hazy saffronBOT
Let me check the knowledge base for you again. One moment! :books:
The knowledge sources only describe the official support channels; they don’t list any “escalation” path beyond what you’re already using, and they don’t mention a way to bypass the AI support bot directly.
From the docs, your options are:
-
Help Center → AI Chat and Phone support
- Use the Help Center and the phone line 1‑888‑GPT‑0090 (where available). [Retiring models]
- The knowledge sources don’t explain how to force a human agent, but the phone option is explicitly listed as a support channel.
-
Be very specific in your support request
Since the AI bot is insisting it’s not 4o mini, include concrete, checkable details:- Your plan: ChatGPT Plus
- Exact model label shown in the UI (e.g., “GPT‑4o mini”)
- Screenshots of:
- The model picker (showing GPT‑5.4 Thinking selected)
- The message header or model label showing 4o mini
- The thinking‑time setting (Standard vs Extended)
- Timestamps and links to affected chats.
-
No other documented support channels
The knowledge sources do not mention email addresses, tickets, or community escalation paths as official support for this kind of account‑specific routing issue, beyond the Help Center chat and phone options. [Retiring models]