[Windows] "Selected model is at capacity" for 6 consecutive days across all models and reasoning efforts (ChatGPT Plus)

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

26.911.7940.0 (OpenAI.Codex_26.911.7940.0_x64__2p2nqsd0c76g0); bundled CLI codex-cli 0.154.0-alpha.6.2

What subscription do you have?

ChatGPT Plus

What platform is your computer?

Windows 11 Pro, Build 26200, x64

What issue are you seeing?

Every normal turn fails with:

Turn error: Selected model is at capacity. Please try a different model.

This is NOT tied to one model or one reasoning effort. On 2026-09-17 alone I logged 52 of these errors in the normal conversation path (codex_core::session::turn).

What I tested today, all failing:

• gpt-5.6-sol, reasoning_effort=high -> "at capacity"
• gpt-5.6-sol, reasoning_effort=low -> "at capacity"
• gpt-5.6-luna, reasoning_effort=high -> "at capacity" (11 consecutive failures in 70 seconds: 16:04:38 -> 16:05:48)

Totals from ~/.codex/logs_2.sqlite (115 failures in codex_core::session::turn over 6 days):
09-07: 2 | 09-08: 3 | 09-11: 23 | 09-12: 16 | 09-16: 35 | 09-17: 36 and counting

Raw log line (PII redacted):

session_loop{thread_id=}:submission_dispatch{otel.name="op.dispatch.turn_input" codex.op="turn_input"}:turn{otel.name="session_task.turn" model=gpt-5.6-luna codex.turn.reasoning_effort=high}:session_task.run:run_turn: Turn error: Selected model is at capacity. Please try a different model.

The retry button re-sends the same model, so it loops on the same error indefinitely.

What steps can reproduce the bug?

1. Run the Codex desktop app on Windows (MSIX build 26.911.7940.0), signed in with a ChatGPT Plus account.
2. Leave the defaults written to ~/.codex/config.toml: model = "gpt-5.6-sol", model_reasoning_effort = "high".
3. Send any prompt in a normal conversation. The turn fails immediately with "Selected model is at capacity. Please try a different model."
4. Switch the model to gpt-5.6-luna, and separately lower model_reasoning_effort to "low". Retry - same error on both.
5. Press the retry/continue button repeatedly - same error, because it re-sends the same saturated model.
6. Repeat on subsequent days - same result every day since 2026-09-07, at all hours of the day.

What is the expected behavior?

A paying ChatGPT Plus subscriber should be able to use Codex, or at minimum:

1. The app should detect that the selected model is saturated and automatically fall back to an available model, instead of failing the whole turn.
2. The retry button should not blindly re-send the same saturated model.
3. The error should be distinguishable from an account-level or plan-level limit, and should carry the request id so support can trace it.
4. There should be a status/incident surface so users can tell a capacity problem apart from a local misconfiguration.

I understand this is a capacity problem on OpenAI's side and I have already opened a support case for it. I am filing this here so there is a public record with concrete data.

Additional information

Environment: Windows 11 Pro Build 26200, ChatGPT.exe MSIX package OpenAI.Codex_26.911.7940.0_x64__2p2nqsd0c76g0, model_provider = "openai", auth_mode = "chatgpt".

Data source: ~/.codex/logs_2.sqlite, table logs, column feedback_log_body, filtered on '%at capacity%'.

Start by reviewing the codex_core::session::turn error path, the retry behavior, and the defaults in ~/.codex/config.toml. Use the reported Windows reproduction and ~/.codex/logs_2.sqlite entries to confirm the failure pattern; done would require an agreed response to capacity errors, retry behavior, and support traceability.