[Bug] "Selected model is at capacity" fires in bursts across all models on an APAC/Japan Pro account #47168

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

26.915.31945 (bundled codex-cli 0.155.0-alpha.9.2)

What subscription do you have?

ChatGPT Pro

What platform is your computer?

macOS (Apple silicon), Japan / APAC

What issue you are seeing?

"Selected model is at capacity. Please try a different model." appears intermittently and in bursts, across multiple models, and repeatedly interrupts the same multi-step task.

Observations:

• Affected models: gpt-5.6-terra (primary), also observed on gpt-6-astra and gpt-5.6-luna.
• It is intermittent rather than permanent. In one measured window (2026-09-21, 07:00-07:49 UTC) roughly 72-75% of gpt-5.6-terra turns failed with an upstream server_is_overloaded condition. Minutes after that window ended, the same model and the same request shape returned HTTP 200 every time (8/8, ~1s each).
• Within a burst the error recurs across many consecutive turns of one task, so a task is only partially completed between failures.

What steps can reproduce the bug?

1. Use the Codex desktop app on macOS with a ChatGPT Pro account from Japan/APAC.
2. Work on a long, multi-step task with gpt-5.6-terra.
3. Observe "Selected model is at capacity. Please try a different model." on some turns; the task continues on other turns.
4. Retry the same turn - the error repeats while the burst is active and disappears once it passes.

What is the expected behavior?

A Codex turn either runs on the selected model, or the failure is reported as a transient upstream capacity error with a retry hint - not as a condition that makes the selected model look permanently unavailable.

Additional information

Things already tried that did not help:

1. Switching models - during a burst every currently offered model is affected.
2. Turning on Fast / priority: service_tier = "priority" in ~/.codex/config.toml, followed by a full quit (Cmd+Q) and relaunch of the desktop app.
3. Full app restart and new tasks.
4. Ruling out a local network problem: outside a burst the identical requests return 200 consistently.

Related context:

• status.openai.com reported "All Systems Operational" with no active incident throughout the same period.
• ⚠ Selected model is at capacity. Please try a different model. #47146 reports the same pattern from another APAC (Thailand) Pro account, where GPT-5.6 Luna additionally disappeared from the model picker. On this machine Luna is still listed.
• Selected model is at capacity. Please try a different model. #47144 already has an OpenAI reply asking for a /feedback session id and the affected models.
• Codex repeatedly fails mid-task with "Selected model is at capacity" across multiple models #47006 asks for an automatic fallback to a compatible model when the selected model is at capacity; there does not appear to be one today.

I can supply a /feedback session id and exact turn timestamps on request.