Summary
When I create a new Codex session in the web UI and explicitly choose a model such as gpt-5.4, the session list/header still shows model: default.
That is confusing for two reasons:
- I can't tell whether the chosen model was actually applied.
default/sonnet/opus looks like a Claude-style mode label, not a concrete Codex model name.
At the same time, the web UI currently exposes no way to choose Codex reasoning / thinking effort.
Repro
- Open the web UI.
- Create a new session.
- Choose agent =
Codex.
- Choose model =
GPT-5.4.
- Start the session.
- Go back to the session list or look at the session header.
Actual behavior
- The UI shows
model: default.
- There is no Codex effort selector.
Expected behavior
- The session list/header should show the actual selected model (for example
gpt-5.4) when one was provided.
- If the UI is showing a Claude-only
modelMode, it should not be presented as if it were the actual model.
- Codex sessions should expose a reasoning effort selector if the backend supports it (
auto/low/medium/high), or the UI should clearly indicate that effort selection is not supported yet.
Why this matters
In a multi-session workflow it becomes impossible to tell which Codex model is running, and the current default label makes it hard to trust what was actually launched.
Summary
When I create a new Codex session in the web UI and explicitly choose a model such as
gpt-5.4, the session list/header still showsmodel: default.That is confusing for two reasons:
default/sonnet/opuslooks like a Claude-style mode label, not a concrete Codex model name.At the same time, the web UI currently exposes no way to choose Codex reasoning / thinking effort.
Repro
Codex.GPT-5.4.Actual behavior
model: default.Expected behavior
gpt-5.4) when one was provided.modelMode, it should not be presented as if it were the actual model.auto/low/medium/high), or the UI should clearly indicate that effort selection is not supported yet.Why this matters
In a multi-session workflow it becomes impossible to tell which Codex model is running, and the current
defaultlabel makes it hard to trust what was actually launched.