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

继续工程师选择继续，或者在新窗口中继续

我继续烂了 编辑器一排继续 字眼出现

好吧，我是自建了个中转反代了我的20pro

收到了回复但是没啥有用的信息

Hello,

Thank you for reaching out to OpenAI Support.

This is Hanna, and I’m taking over your case regarding the persistent “Selected model is at capacity. Please try a different model.” errors you have been encountering in Codex CLI.

We understand that this has been occurring repeatedly since August 11 across multiple models, including GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna, GPT-5.5, and Codex Auto Review, with a separate stream-disconnect error also reported for GPT-5.3-Codex-Spark. We know how disruptive repeated failures like these can be when you are trying to use Codex through your Pro account.

We have reviewed the diagnostic information you already provided. You do not need to resend your request IDs, timestamps, model names, error messages, account information, Pro status, concurrency information, or VPN/proxy details.

To clarify an important point, the reported “Selected model is at capacity” message is not, by itself, an indication that your Pro entitlement or available Codex usage has been exhausted.

Codex usage and credits are metered according to the actual processing performed. Under the current token-based rate card, consumption depends on the model and the combination of input, cached input, and output tokens processed. Different models therefore consume usage at different rates.

For example:

• GPT-5.6 Sol: 125 credits per 1M input tokens, 12.5 credits per 1M cached input tokens, and 750 credits per 1M output tokens.


• GPT-5.6 Terra: 50 / 5 / 300 credits respectively.


• GPT-5.6 Luna: 5 / 0.5 / 30 credits respectively.


• GPT-5.5: 125 / 12.5 / 750 credits respectively.

Having available Pro usage or credits does not guarantee that every model will be available for every request at a particular moment. Model availability and credit consumption are separate considerations.

In your case, the diagnostic context is particularly relevant because you have confirmed that:

• the failures occur through Codex CLI;


• web/desktop do not reproduce the same capacity behavior;


• concurrency is limited to one run;


• no VPN or proxy is being used; and


• the errors have occurred across several models and multiple dates.

You also provided multiple request IDs showing the capacity condition as well as the separate stream-disconnect event. We have those details already, so there is no need to generate or submit additional request IDs solely to repeat information already captured in this case.

For additional information about how Codex usage is calculated, please see: Codex Rate Card

We understand that you specifically requested direct human assistance rather than another cycle of automated questions. Your case information has been preserved for this review, and we have avoided asking you to repeat diagnostics that are already documented.

If you have any other questions or concerns, please let us know. We’re here to help.

Best,

Hanna

OpenAI Support