Codex App subagents appear to use Fast mode despite Standard speed and fast opt-out #30407

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

26.623.42026 (build 4514)

What subscription do you have?

ChatGPT Pro

What platform is your computer?

Darwin 25.5.0 arm64 arm

What issue are you seeing?

I am using the Codex App on macOS, not the CLI. The main session Speed is set/displayed as Standard, and I have not intentionally enabled Fast mode.

However, Fast usage appears in the usage breakdown after running dynamically spawned subagents from the Codex App. My workflow does not define subagents in TOML. Instead, I ask the main agent in natural language to create worker subagents, with gpt-5.5 low reasoning / "5.5 Light" for the subagents.

Relevant local config:

model = "gpt-5.5"
model_reasoning_effort = "medium"
service_tier = "default"

[notice]
fast_default_opt_out = true

There is no fast_mode entry under [features] in my config.

I could not find a way for the Codex App's dynamic subagent spawn flow to explicitly pass or force service_tier = "default" / Standard for a subagent. Natural-language instructions such as "Fast mode off" or "use Standard" do not appear to provide an enforceable tier setting.

I cannot prove from the UI alone that every Fast token came from subagents, but the behavior appears inconsistent with the main session being Standard and the config opting out of Fast defaults.

What steps can reproduce the bug?

1. Use Codex App version 26.623.42026 (build 4514) on macOS.
2. Configure ~/.codex/config.toml with service_tier = "default" and [notice] fast_default_opt_out = true. Do not set [features].fast_mode.
3. Start a main Codex App session with model gpt-5.5, reasoning effort medium, and Speed shown as Standard.
4. Ask the main agent to dynamically spawn worker subagents using gpt-5.5 low reasoning / "5.5 Light".
5. Ask that the subagents run with Fast mode off / Standard behavior.
6. Check the usage breakdown after the subagents run.

Observed result: Fast usage appears despite the main session being Standard and the config opting out of Fast defaults.

Closest issue I found before drafting this: #29940. That issue is similar, but it reports CLI/Linux behavior. This report is specifically for the Codex App on macOS.

What is the expected behavior?

Spawned subagents should respect the effective service tier of the main Codex App session/config by default.

If the main session is Standard and the config uses service_tier = "default" plus fast_default_opt_out = true, dynamically spawned subagents should not silently use Fast mode.

If subagents intentionally use a different service tier, the app should show that clearly and provide a way to force Standard/default behavior for subagents.

Additional information

This may be related to #29940, but that report is for the CLI. The issue here is the Codex App dynamic subagent flow on macOS.

Possible cause: service tier may be resolved independently for app-spawned subagents, or the app UI may not expose the effective tier used by subagents.