Feedback Regarding Codex Model Availability Issue

ChatGPT Bugs
1

Hello OpenAI ,

I am a Pro 20x user, and I would like to report a very frustrating issue with Codex.

Every time I try to use Codex, I receive the following error message:

“Selected model is at capacity. Please try a different model.”

This has happened repeatedly, and it prevents me from using Codex normally. I subscribed to the Pro 20x plan because I need reliable access to OpenAI’s advanced tools, especially Codex for my development work. However, constantly being blocked by capacity limitations makes me feel very disappointed and frustrated.

I understand that high demand can sometimes affect availability, but as a paying Pro user, I expected a more stable experience and better access priority.

This situation makes me question the value of my subscription, because the feature I need most is unavailable when I try to use it.

Please investigate this issue and improve Codex availability for Pro users. If there are temporary capacity problems, I hope OpenAI can provide better communication or prioritize paid subscribers.

I really enjoy OpenAI products and want to continue using them, but this experience has been very disappointing.

2

Hello OpenAI Team,

I am following up on my previous feedback regarding the Codex capacity issue.

Although the “Selected model is at capacity” problem appears to have improved, the actual user experience has not been restored. The problem has changed from “unable to access the model” to “able to access the model but receiving a significantly degraded experience.”

As a Pro 20x user, I am extremely disappointed to find that even when I select 6Astra, the model performance no longer feels like the 6Astra experience I subscribed for. The reasoning ability, understanding of complex tasks, and overall response quality feel much closer to GPT-4o-level performance rather than the expected 6Astra capability.

This makes me strongly suspect that there may be a model routing issue, where 6Astra requests are being routed to GPT-4o or another lower-capability fallback model, possibly due to capacity management or backend resource allocation.

If OpenAI is making routing changes or using fallback models during periods of high demand, paying users deserve transparency. A Pro 20x subscription should not silently become a lower-tier experience without any notification.

The most frustrating part is that the issue is not only about availability anymore. The model is technically accessible, but the intelligence, accuracy, and responsiveness have noticeably declined. This means the core value of the subscription is still not being delivered.

I have attached an example of the degraded output quality. The generated result demonstrates the type of quality regression I am experiencing.

I sincerely hope OpenAI investigates whether:

  • 6Astra is being correctly served to users who select it
  • Pro users are being silently routed to lower-capability models
  • Current capacity management is causing hidden model degradation

I upgraded to Pro 20x because I trusted OpenAI to provide the best available experience. Right now, I feel disappointed because the service I am receiving does not match what I paid for.

Please investigate this issue and provide a clear explanation.

Thank you.

3

pelican_bicycle
pelican_bicycle1000×625 8.09 KB

4

LOL , this picture is really crazy.

5

Yes, this is my conclusion as well, after experiencing a sudden substantial (to the point of now being unusable for my work) loss in image quality - especially noticeable in photorealistic lighting and rendering - for the past six days or so, which lines up with the release of Astra.

It simply seems like every single request is being funnelled to a more simplified model or process, regardless of whether I choose Astra, generate in work mode, effort or thinking level, etc.

Not only has the quality dropped, but every prompt reproduces essentially the exact same low quality image every time.

Six days now, still no difference. Obviously lots of other users are having their own issues with Astra - including this exact one - so I’m holding out to see if things begin to settle in the coming days.

However, as of writing this, the status page has been updated to acknowledge failures for image generation. Whether this touches on the same issue here - as I would consider the results of these generations as failures in functionality - we’ll see. Will update if I notice changes after the status page cliams to have rolled out a fix.

6

Hello OpenAI Team,

I would like to provide a follow-up feedback regarding my previous reports about model availability and quality issues.

After finally gaining access to the 6Astra model and using it for about one day, I would like to share my actual experience.

At first, I was very excited to try 6Astra because, as a Pro 20x user, I expected this model to provide a significant improvement in reasoning ability, coding assistance, and overall intelligence.

However, after a full day of usage, I found that the current experience still has many issues.

The first problem is performance speed. The model feels much slower than expected. Even simple requests can take a long time to process, which makes daily usage inefficient. After waiting through previous capacity issues, it is disappointing to finally access the model but still experience slow responses.

The second problem is the overly aggressive cyber-related restrictions. During normal usage, many legitimate technical questions are unexpectedly classified as cyber-related. These requests are often about programming, debugging, learning concepts, or general technical assistance, but the model frequently blocks or limits the conversation.

After using 6Astra for a day, my impression is that the model has strong potential, but the current balance between safety and usability is not ideal. The system seems to focus too much on restriction and not enough on understanding user intent.

For Pro 20x users who rely on AI for productivity, this creates a frustrating experience. Advanced users naturally ask more technical questions, and excessive false positives significantly reduce the usefulness of the model.

This is another issue following my previous feedback:

  • First, users experienced difficulty accessing the model due to capacity limitations.
  • Then, after access was restored, there were concerns about model quality and performance.
  • Now, after trying 6Astra itself, the experience is affected by slow speed and excessive restrictions.

I hope OpenAI can continue improving 6Astra by:

  • Improving response speed and stability
  • Reducing unnecessary cyber-related false positives
  • Better understanding legitimate user intent
  • Providing Pro users with the level of experience expected from a premium subscription

I believe 6Astra has great potential, but the current experience still feels limited compared with what Pro 20x users expect.

Thank you for taking this feedback seriously.

7

Hello OpenAI ,

I am following up on my previous reports about 6Astra. I am a Pro 20x subscriber who paid $200, and I am honestly at the end of my patience.

I have already reported repeatedly encountering this error:

“Selected model is at capacity. Please try a different model.”

Now I am experiencing something that makes this situation even more frustrating: my requests keep failing to go through when I write in Chinese, while requests written in English can go through.

Why should the language I use appear to determine whether I can access the model I paid for? Why am I having to switch to English just to get a response?

Are Chinese-language requests being treated differently? Are Chinese-speaking users receiving a worse experience simply because of the language they use?

I cannot verify what is happening behind the scenes, so I am not claiming to know the cause. But this is the difference I am experiencing, and it raises a serious concern about whether access is being handled consistently across languages.

I paid for access to 6Astra. I did not pay to keep guessing whether a request will work, which language I need to use, or whether I will run into another capacity error. Using Chinese should not make me feel like a second-class customer.

My previous post was hidden following community flags, but the problems I was reporting have not gone away. Having 6Astra listed in the model selector means very little to me when I cannot actually use it reliably in the language I normally work in.

After all the capacity problems, slow responses, and inconsistent quality, I now have to deal with this unexplained difference between Chinese and English as well.

I spent $200 expecting a service I could actually use. This is not what I paid for. I am angry, exhausted, and completely fed up. I really cannot take this anymore.

8

Note: I am not an OpenAI employee.

The only mention of Codex is in the title, so I am assuming this issue occurs when the prompts are being sent through Codex using GPT-6 Astra.

I searched the OpenAI Codex issue tracker for an existing issue that closely matches the behavior you described with Astra and Chinese-language prompts, but I could not find one close enough that I would recommend reading it.

As such, I think it would be beneficial to open a new issue in the Codex issue tracker.

Include as much relevant information as possible, such as your Codex version, operating system, environment, the model used (GPT-6 Astra), steps to reproduce, expected behavior, actual behavior, and any relevant logs or error messages. It would also be useful to include example prompts that work in English but fail when equivalent prompts are submitted in Chinese.

Also consider including a Feedback ID, which may provide OpenAI with additional diagnostic information: