5.6 pro model has been automatically downgraded and routed to the 5.5 mini model since its release

ChatGPT Bugs
1

Hi everyone,
I am a subscriber of the ChatGPT Pro 5x tier.

Since the release of version 5.6, I haven’t had the chance to use the Pro tier model. I can use the Thinking model normally, but the Pro model will be automatically routed down to the 5.5-mini model:

pro model
pro model1421×1321 103 KB

thinking model
thinking model1431×1033 69.6 KB

In the HAR network log, it can be seen that my subscription level information and requests are all normal, but the returned model is a downgrade model:
“plan_type”:“prolite”

“model_slug”:“gpt-5-6-pro”,“default_model_slug”:“gpt-5-6-pro”,“thinking_effort”:“standard”

“resolved_model_slug”:“gpt-5-5-mini”

Therefore, I am quite certain that there must be an abnormal issue with my account on the server side. I have sent a screen recording and the har log file from the operation at the same time to the help center to reproduce the error phenomenon. However, I have not received any feedback from the human experts yet, and the Codex quota is not as durable as before. I am not sure about the differences between my current account and the Plus subscription. I wonder if others have the same problem as me.

The case number I have received is 11851745.

I hope my account issue can be resolved and replied to. At least, I would like to know the current handling progress instead of waiting indefinitely.

2

same as you. I even didn’t use pro model, just use “medium” level and it said that “5.5 mini”, not 5.6 thinking days before.

3

I hope this issue can be resolved as soon as possible. Otherwise, I don’t know what the point of subscribing to the Pro tier account is at present :frowning:

4

One day has passed and it’s almost time to leave work. Still, there is no news or progress at all :smiling_face_with_tear:

5

The same issue, 1 hour ago.

6

Me too for these days. I feel really disappointed for the degraded Pro’s service.

7

Exactly same situation with 20x. Any solutions so far?

8

Exactly the same situation, never used Pro model since the last update.

9

Since the launch of version 5.6, I have been experiencing the issue where the 5.6 Pro model is automatically downgraded and routed to the 5.5 mini model.

At the beginning, the issue would occasionally resolve itself. Sometimes the Pro model would work normally on one platform (mobile app, desktop app, or web version) while remaining abnormal on the others. However, during the past week, the issue has become almost constant. Only on very rare occasions has the Pro model worked normally on my mobile app.

I started reporting this issue to customer support on July 10, but I have never received any real assistance or a meaningful solution. It feels like OpenAI does not actually want to resolve this problem and has only been providing superficial responses. As a result, the issue remains unresolved to this day.

I am a long-time ChatGPT user. I have been subscribing to ChatGPT continuously since 2022. However, in order to regain normal access to the Pro model, I reluctantly abandoned my four-year-old account, canceled the subscription, and created a new account to subscribe again.

Then an unexpected situation occurred. I subscribed to Pro 5x on two accounts through the App Store. Among them, only one account was able to use the Pro model normally on the mobile app twice before receiving a usage limit notification. I suspect this happened because previous requests that were automatically routed from the Pro model to the mini model were still counted toward my Pro model usage quota.

I then suspected that the issue might be related to the payment method. Therefore, I asked a friend to help me subscribe using a Visa card and purchased another Pro 20x plan. However, the exact same problem occurred again. Even after changing devices and networks, logging into my account still resulted in the same routing issue.

I do not understand why this problem only affects my accounts. Could it be related to the fact that all three of my accounts have been logged into the ChatGPT mobile app, or that they are all associated with Gmail? What makes this even more confusing is that when I logged into my friend’s account on the same device, I was able to use the Pro model normally for five consecutive days. The usage frequency on my friend’s account was actually much higher than on my own account, while I sometimes only use my own account once every two or three days.

This month, in an attempt to get the Pro model working normally, I have spent a total of $500 on Pro subscriptions across different accounts. In return, I have only lost time, energy, and money, while still being unable to properly use the service I paid for.

I sincerely hope that someone can investigate this issue and help me resolve it.
The case number I have received is 11240087

10

Adding another reproducible case to this thread.
Case Number: 12153385
I am a paid ChatGPT Pro 200$ subscriber. When I manually select GPT-5.6 Pro, the server response metadata shows:
model_slug: gpt-5-6-pro
default_model_slug: gpt-5-6-pro
resolved_model_slug: gpt-5-5-mini
OpenAI Support has acknowledged the discrepancy between the premium model selected in the interface and the lower-capability model shown in the server metadata. My case has now been escalated to a support specialist, with further investigation requested from the routing, entitlement, and billing teams.

This now requires a clear written explanation from OpenAI:
What exactly does resolved_model_slug represent?
Were the affected requests actually processed by GPT-5.5 mini?
Why did the interface continue to display GPT-5.6 Pro without any downgrade, fallback, usage-limit, or reset-time notice?
Were these downgraded requests counted against the GPT-5.6 Pro usage allowance?
How many paid subscribers are affected, and when did this issue begin?
What refund, credit, or other remediation will be provided for the affected subscription period?

This is a paid-service transparency and entitlement issue. A user who explicitly selects and pays for a Pro model should not have to inspect HAR files and undocumented server metadata to determine which model was actually delivered.

Please provide:
1.confirmation that this issue is being tracked centrally;
2.an incident or internal tracking reference;
3.an estimated investigation or resolution timeline;
4.a public explanation of the applicable routing and fallback policy.

11

Adding another reproducible cross-platform case to this thread.

Support case numbers:

  • 10850343
  • 11030523
  • 11605440
  • 11930987

I am a long-term ChatGPT user and originally subscribed to the USD 100/month Pro 5x plan through Apple. I later upgraded the same account to the USD 200/month Pro 20x plan specifically to test whether the higher allowance or a refreshed subscription entitlement would resolve the issue.

It did not. The Pro-to-mini behavior remained unchanged immediately after upgrading to Pro 20x.

The issue has affected both Windows web and the iOS app.

Main symptoms:

  • GPT-5.5 Pro Extended and GPT-5.6 Pro remain selectable in the model picker.
  • No usage-limit warning is displayed.
  • No reset timer or temporary-unavailability notice is displayed.
  • Pro responses frequently begin almost immediately, with no meaningful Pro-level reasoning period and materially degraded reasoning quality.
  • GPT-5.6 Extra High continues to reason normally and usually produces substantially stronger results.
  • The behavior sometimes temporarily recovers on iOS, but later returns on the same device and account without any Windows usage.
  • Existing conversations and newly created conversations can behave differently, but new conversations are also affected.

Previous HAR evidence and Support review documented a requested-versus-resolved model mismatch. The Pro model was selected with the expected reasoning setting, while the backend metadata resolved the response to a mini model, including gpt-5-3-mini and later gpt-5-5-mini.

I also conducted a controlled behavioral comparison after upgrading to Pro 20x:

  • The exact same complex, closed-book prompt was submitted in five newly created conversations using Instant, Medium, High, Extra High, and Pro.
  • The five answers were evaluated blindly before the mode labels were disclosed.
  • The ranking was:
    1. Extra High
    2. High
    3. Medium
    4. Instant
    5. Pro

The Pro answer ranked last, below even Instant, and contained substantive reasoning and compliance errors that were not present in the Extra High and High answers. I understand that output quality alone does not establish the exact backend model, but it is consistent with the prior HAR routing evidence.

Troubleshooting already completed:

  • Chrome and Edge
  • Normal and incognito browser sessions
  • Cleared ChatGPT site data and cookies
  • Logged out of all devices and logged back in
  • Password reset and 2FA
  • Windows web and iOS
  • Clash and v2rayN
  • TUN and system-proxy modes
  • Fixed Singapore, Japan, and United States nodes
  • Multiple HAR files
  • Screen recordings, screenshots, timestamps, conversation links, request IDs, and trace information
  • Apple “Restore Purchases”
  • Pro 5x to Pro 20x upgrade

None of these produced a stable resolution.

Support has repeatedly acknowledged the reported inconsistency and the requested-versus-resolved metadata, but has not confirmed an active engineering investigation or provided a backend diagnosis. Multiple support cases have been created and reassigned, while the same diagnostic evidence has repeatedly been requested.

Recent affected conversation details, exact timestamps, conversation URLs, screenshots, recordings, and HAR files are already available privately in the support cases listed above.

@OpenAI_Support, please add Cases 10850343, 11030523, 11605440, and 11930987 to the centrally tracked group of related Pro-routing reports.

Please confirm:

  1. Whether this issue is being centrally tracked;
  2. Whether the affected cases have been linked;
  3. Whether the routing, entitlement, allowance, or account-state systems are being investigated;
  4. Whether downgraded turns are counted against the paid Pro allowance;
  5. Whether an incident reference or investigation timeline can be shared.

I do not want to leave ChatGPT or request a refund. I want the Pro service I paid for to work correctly and consistently.

12

Adding another reproducible case from July 31, 2026 (ChatGPT web in Chrome).

I am a paid ChatGPT Pro subscriber. I manually selected Pro in the model picker. The response itself identified the current instance as GPT-5.5-mini, and the conversation response metadata showed:

resolved_model_slug: gpt-5-5-mini

The UI continued to display Pro, with no usage-limit warning, fallback notification, or reset timer. This remained reproducible after the official service status returned to normal.

This appears to be the same requested-versus-resolved routing mismatch reported in this thread. @OpenAI_Support, please link this report to the central routing/entitlement investigation and clarify:

  1. Whether the affected requests were actually processed by GPT-5.5-mini;
  2. Whether they were counted against the paid Pro allowance;
  3. Whether an account-specific remediation or incident reference is available.

The attached screenshot is cropped to show the selected Pro mode and the model response; private account and conversation data have been excluded.

chatgpt-pro-mini-evidence-cropped
chatgpt-pro-mini-evidence-cropped1050×790 64.7 KB

13

image
image783×335 6.95 KB

Still happening… Thought I was going crazy.

14

There must be legislation to ensure consistency between the actual AI models being used and what is shown on the front-end interface. AI vendors are cutting corners and playing deceptive tricks, and this kind of malpractice must be resolutely cracked down on.

15

Happend to me as well. This is annoying

16

Happened to me as well. I made a full GPT 5X subscription on google play on July 29th and the 5.6 pro model works till July 31st, when I try to use it today, it down to 5.5 mini, quite annoying.

17

My friend’s account and mine developed this issue one after another, but official support has made no real effort to resolve it. I first reported the problem on July 10, and it still has not been fixed as of today.

When you report the issue to official support, they escalate the case to a human support agent. The agent then says that the problem will be forwarded to the technical team and asks you to submit a screen recording. However, even after you upload the recording—and even when the SSE logs clearly show an obvious model-routing error—they still do not resolve the issue, and there is no further follow-up.

A few days later, when you ask for an update, you receive another AI-generated response. You then have to request escalation to a human agent again. The case is once again forwarded to the technical team, you are asked to record and upload another video, and then there is no response. I have already gone through this entire process twice.

So you may as well give up. This problem appears to be impossible to resolve. I even suspect that OpenAI has deliberately made this adjustment because it has diverted computing resources to the development of new models. Recently, more and more of my friends have encountered the same issue, but OpenAI simply does not seem willing to fix it.

18

Everyone needs to talk about this so the noise gets past this little corner no one sees unless they’ve hit the same problem.

19

For more than 2 weeks I’ve been dealing with this exact routing error. I haven’t even been using pro model ever, just “medium” level. But it’s still 5.5mini answering my question. As a pro subscriber, I feel very disappointed facing this problem

20

Happening to me too as a 20x subscriber, super annoying