Codex Desktop reasoning level resets from xhigh/high to low in the same active thread after delegations/continuations

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

Codex App 26.602.40724 (CFBundleVersion 3593) from /Applications/Codex.app.

What subscription do you have?

ChatGPT Pro subscription, USD 200/month.

What platform is your computer?

Darwin 25.5.0 arm64 arm

What issue are you seeing?

In Codex Desktop, the reasoning/thinking level keeps resetting to low inside the same active thread, even after I manually set it to high / xhigh.

This is not only a new-chat/default-setting issue. It happens during an ongoing operational thread with active delegated Codex threads/workers. I set the level to extremely high, continue working, and shortly after a delegation/update/continuation it shows as low again without me changing it.

Why this matters: I use Codex as an operational coordinator across multiple live project threads. Some tasks involve production systems, SQL/Supabase, Vercel/env, auth/secrets, RRHH/payroll, cash/stock/sales/accounting, and push/main decisions. For these workflows, reasoning effort must not silently drop to low, because the user may believe the thread is still running with high/xhigh reasoning.

What steps can reproduce the bug?

1. Open a Codex Desktop thread.
2. Set reasoning/thinking level to xhigh or high.
3. Use the thread as a coordinator and send/receive codex_delegation messages to/from other active Codex threads.
4. Continue working while background captain/worker threads are active.
5. Observe that the same thread’s reasoning level changes back to low without the user selecting low.

Active thread IDs at time of issue:

Current/active coordination thread:

• 019ea4f4-7990-7e32-a582-8ae8c3640303 — Coordinar capitanía general

Active captain/worker threads involved:

• 019e8074-ed13-7193-be78-eae91235c9f9 — Capitán - Avícola del Sur
• 019ea4ea-e759-7290-9579-11f5b206913b — Worker - Avícola producción viva QA integral no-RRHH
• 019e8081-6810-70d2-a27e-40d8c5697024 — Capitán - Turilo ERP
• 019ea4ea-fa86-7af0-b0ee-bca2cf822480 — Worker - Turilo producción viva entrega mañana

Related/historical CAPITAN GENERAL thread used by delegations:

• 019e7c7d-b636-7fe0-98a8-e20dab577fb1 — CAPITAN GENERAL

What is the expected behavior?

If I manually set reasoning to xhigh or high in an active Codex Desktop thread, it should remain at that level unless I explicitly change it.

At minimum, Codex should not silently reset the visible/current reasoning effort to low during the same thread after:

• thread continuation,
• context compaction,
• receiving codex_delegation messages,
• reading/sending messages to background threads,
• active worker/captain updates,
• or returning from tool activity.

Additional information

This may be related to thread rehydration, continuation, context compaction, or delegation handling resetting the active mode settings.

This is similar in spirit to existing reasoning-effort persistence issues, but the observed bug is specifically:

• Codex Desktop
• same active thread
• user manually set high/xhigh
• resets to low later without user action
• during delegation/background-thread workflows

Related issues:

• #12042 — Codex CLI resets model_reasoning_effort
• #21093 — Codex Desktop new-chat reasoning selector does not apply to initial prompt
• #13536 — Codex app automations run at medium reasoning even when global effort is xhigh
• #26718 — resume_agent restores subagent history but not original model/reasoning settings
• #25583 — Codex iOS remote task on Windows 26.527 ignores xhigh and starts gpt-5.5 medium