[Codex Desktop] Astra returns invalid_prompt; Sol, Terra, and Luna frequently return capacity errors

What version of the Codex App are you using (From “About Codex” dialog)?

26.901.51231

What subscription do you have?

Pro 20x

What platform is your computer?

Microsoft Windows NT 10.0.26200.0 x64

What issue are you seeing?

I am experiencing two types of failures in Codex:

1. GPT-6 Astra consistently fails with invalid_prompt.

Exact error message:

Invalid prompt: your prompt was flagged as potentially violating our usage policy. Please try again with a different prompt: https://platform.openai.com/docs/guides/reasoning#advice-on-prompting

This occurs with low, high, and max reasoning effort. The issue has persisted since approximately 2026-09-07 00:00 UTC and remains reproducible as of 2026-09-08.

2. GPT-5.6 Sol, Terra, and Luna frequently fail with a capacity error.

Exact error message:

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

This happens across all tested reasoning effort levels. Switching models or changing reasoning effort does not reliably resolve the problem, because the alternative models also frequently return the same capacity error.

These capacity errors are frequent, but I am not claiming that every request to these models fails.

Product / Model	Reasoning effort	Observed behavior
Codex — GPT-6 Astra	low, high, max	Consistently fails with invalid_prompt
Codex — GPT-5.6 Sol	All tested levels	Frequently returns the capacity error
Codex — GPT-5.6 Terra	All tested levels	Frequently returns the capacity error
Codex — GPT-5.6 Luna	All tested levels	Frequently returns the capacity error
Regular ChatGPT web chat — GPT-6 Pro	—	Works normally

What steps can reproduce the bug?

On the affected account:

1. Open Codex and start a new conversation.
2. Select gpt-6-astra and send a message.
3. Observe the invalid_prompt error.
4. Repeat with low, high, and max reasoning effort. The Astra failure persists.
5. Switch to GPT-5.6 Sol, Terra, or Luna and send a message.
6. Frequently receive:
   Selected model is at capacity. Please try a different model.
7. Change the reasoning effort or switch among these models. Capacity errors continue to occur frequently.

For the Astra invalid_prompt issue, I have already confirmed that it persists across:

• Multiple devices.
• New conversations.
• Clean configurations.

These checks have not resolved the Astra failure. I have not yet established whether the two error types share the same underlying cause.

What is the expected behavior?

Valid requests should work normally when model access and capacity are available.

If a model is temporarily unavailable, Codex should provide actionable guidance. The current suggestion to switch models is not a reliable workaround because Sol, Terra, and Luna also frequently report capacity errors.

If an account-level or model-specific restriction applies, the error should clearly explain the restriction and the appropriate next steps.

Additional information

Related Astra invalid_prompt reports:

• #43058
• #43163
• #43237

Related capacity-error reports:

• #43337
• #43368

I am referencing these reports because the symptoms overlap, not because I have confirmed a shared root cause.

The combination of persistent Astra failures and frequent capacity errors across the other models makes Codex unreliable for normal use. Sol should not be considered an unaffected or reliable workaround in this case.

I suspect a possible account-related restriction or backend issue, but I cannot verify this from the client. Could the team correlate the affected requests with backend logs and investigate:

• Why Astra consistently returns invalid_prompt.
• Why Sol, Terra, and Luna frequently return capacity errors across reasoning effort levels.
• Whether these failures reflect actual model capacity, account-level access or throttling, safety classification, routing, or separate issues.

session ID:
[01a06553-317c-7711-96b6-b248ac8f2f8a]
[01a07e67-3d21-7051-a50d-2efb78f8697f]

Start by reproducing the Astra and Sol, Terra, and Luna failures using the listed model selections, reasoning levels, and session IDs. Correlate those requests with backend logs to determine whether the failures are caused by capacity, account access, throttling, safety classification, or separate issues; done means identifying the cause and defining actionable client guidance.