Codex Says "Selected Model Is at Capacity"? Here's What Users Report and What to Try

What the error looks like

The full message reads "Selected model is at capacity. Please try a different model." A search of the openai/codex repository at 20:44 Beijing time on October 20 found 24 issues opened since October 18 that match "at capacity," including 14 opened on October 20 (UTC). Not all of those are about this error, but more than a dozen have it in the title. The reports cover the Codex CLI, the desktop app and Codex on the web.

The reports describe two different patterns. In issue 55544, filed at 19:28 Beijing time on October 20, a ChatGPT Plus user with gpt-6-astra set as the model says about 30% of requests fail with the message and that "an immediate retry usually succeeds," so the failures are intermittent. The user Cartoncase says the error also appeared on several other models, including gpt-6-luna, gpt-6-sol and gpt-6.1-sol, and in the desktop app. Cartoncase later closed the issue as a duplicate.

In issue 55470, filed at 14:04 Beijing time on October 20, a ChatGPT Pro user says every request to GPT-6.1 Sol and GPT-6 Sol was rejected for more than seven hours on October 19, Pacific time, across the CLI, the desktop app and the web. The user WayneChenDG says "Retry does not help" and that the Codex status command showed "Weekly limit: 78% left" at the time.

What has helped

Based on these reports, the first thing to try depends on the pattern. If the error comes and goes, issue 55544 says an immediate retry usually succeeds. If it keeps failing on one model, the error message itself suggests trying a different model, and issue 55470 says starting a new chat and selecting GPT-6 Luna worked. Switching models did not help everyone, though, since issue 55544 lists GPT-6 Luna among the affected models.

DigIntoAI found no OpenAI documentation that explains this error, so these steps come from users' reports rather than from OpenAI.

What OpenAI has said

OpenAI has not explained the cause. In the issues DigIntoAI checked, the only replies other than those from users were automated duplicate notices, and no OpenAI staff member had responded. As of 20:44 Beijing time on October 20, the OpenAI status page showed "All Systems Operational" with no active incident.