[App] create_thread advertises gpt-5.6-sol/xhigh but rejects it during validation

What version of the Codex App are you using?

Codex App 26.715.7063.0

What subscription do you have?

Not included in this report.

What platform is your computer?

Windows 11 Pro 10.0.26200 x64

What issue are you seeing?

The first-party codex_app.create_thread tool description on the calling host explicitly advertises this combination:

• model: gpt-5.6-sol
• reasoning effort: xhigh

However, a single capability-only create attempt with exactly that model/effort pair was rejected before thread creation with:

create_thread could not validate reasoning effort "xhigh" for model "gpt-5.6-sol". Use a model and reasoning combination listed in the tool description, or omit thinking.

This makes the operation-specific capability contract self-contradictory: the exposed tool schema/description says the pair is supported, while the create-path validator says it is not.

The observation was made at 2026-07-19T15:18:09Z. After the rejection, an exact read-only lookup using a unique correlation token found zero created threads. The request was not retried.

What steps can reproduce the bug?

1. Start a Codex Desktop thread where the codex_app.create_thread tool description lists gpt-5.6-sol with supported reasoning efforts including xhigh.
2. Invoke create_thread once with:

• a projectless target
• model: "gpt-5.6-sol"
• thinking: "xhigh"

1. Observe that the create-path validator rejects the same combination that the tool contract advertised.

What is the expected behavior?

The advertised contract and the create validator should derive from one versioned capability authority.

Either:

• gpt-5.6-sol/xhigh is accepted by create_thread; or
• xhigh is omitted from the advertised gpt-5.6-sol create capability.

On success, the create response should include enough operation-specific evidence to identify the actual selected model and reasoning effort. Ambiguous responses should also be reconcilable without requiring a retry.

Additional information

Only one create attempt was performed. No benchmark, duplicate create, model mutation, or session mutation was run.