[Codex App][Voice Chat] Unsupported minimal reasoning effort causes request failure

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

26.721.4979.0

What subscription do you have?

Not disclosed (not relevant to reproduction).

What platform is your computer?

Microsoft Windows NT 10.0.26200.0 x64

What issue are you seeing?

Starting or continuing a Voice Chat can fail because the app sends reasoning_effort=minimal to the voice-selected model, while that model rejects minimal and supports only none, low, medium, high, and xhigh.

Redacted error text:

Unsupported value: 'minimal' is not supported with the selected realtime voice model. Supported values are: 'none', 'low', 'medium', 'high', and 'xhigh'.

The failure was observed in two separate Voice Chat threads. Sending a subsequent turn with reasoning_effort=medium succeeded and restored both affected threads.

What steps can reproduce the bug?

1. Open the Codex App on Windows.
2. Start a new Voice Chat.
3. Send a short utterance after the voice session starts.
4. Observe that the request can fail with the unsupported minimal reasoning-effort error.
5. Start another Voice Chat; the same error may recur.

The behavior may depend on the new-thread defaults or an active feature treatment, so reproduction might not be universal.

What is the expected behavior?

Before starting a realtime voice turn, the app should intersect the requested reasoning effort with the active model's supported values. It should never send minimal to a model that does not advertise support for it.

A safe fallback such as medium, or normalization based on model capabilities, would prevent the request from failing.

Additional information

• The user-level Codex configuration was valid and used model_reasoning_effort = "xhigh".
• codex doctor --json reported healthy configuration and state checks.
• A manual medium turn succeeded immediately, indicating a reasoning-effort compatibility mismatch rather than a general Voice Chat outage.
• Inference: a new-thread Voice Chat default may be applied without validating it against the selected model's supported reasoning efforts.
• This is intentionally a text-only public report. No desktop logs, screenshots, session/thread IDs, local filesystem paths, account details, or complete internal model identifiers are attached.