[IDE] Max reasoning effort is missing in the VS Code extension while available in Codex App

[akitin-work]

What version of the Codex IDE extension are you using?

openai.chatgpt-26.721.41059-win32-x64
Bundled Codex CLI: 0.146.0-alpha.3.1

Platform

Windows, VS Code extension.

What issue are you seeing?

The Codex App Work surface exposes the Max reasoning effort for GPT-5.6-Sol and it is usable there, but the VS Code extension does not show Max in its reasoning-effort picker.

This is not a request to enable Max in the Codex App; Max is already present and working there. The missing option is specifically in the VS Code extension.

Local configuration already includes:

[desktop]
enabled-reasoning-efforts = ["low", "medium", "high", "xhigh", "ultra", "max"]

The active model catalog also advertises max support for gpt-5.6-sol. Reloading VS Code and starting a new chat did not make the option appear.

Steps to reproduce

1. Open the Codex VS Code extension.
2. Select GPT-5.6-Sol.
3. Open the reasoning-effort picker.
4. Observe that Max is absent.
5. Open Codex App Work with the same account/model.
6. Observe that Max is available and usable there.

Expected behavior

For the same account and model, the VS Code extension should expose every officially enabled reasoning effort, including Max, or clearly document why the IDE surface intentionally differs from Codex App Work.

Actual behavior

The Codex App and VS Code extension expose different reasoning-effort choices even though both the local configuration and the model catalog include max.

Additional feedback

Troubleshooting through Codex also made this worse: the agent first suggested that reloading or a hidden/advanced setting would expose Max, then claimed a local UI fix was complete based only on static bundle checks. After reloading VS Code, Max was still absent. A Codex agent should not report an IDE UI issue as fixed without verifying the actual rendered UI.

This issue describes the original behavior before that unsuccessful local experiment. No personal information, account identifiers, local filesystem paths, or private conversation content are included here.

[sirouk]

@akitin-work you can do this:

[akitin-work]

Thanks for sharing this — your screenshot helped me confirm that I was looking at the relevant setting. I confirmed that all six reasoning efforts, including Max, are already enabled on my side. However, the VS Code extension's picker still shows Light, Medium, High, Extra High, and Ultra, but not Max. It seems that Max is enabled in the configuration but is not appearing in the VS Code extension's picker.

[mtamburrano]

I have the same issue.
Alle reasoning efforts selected in the codex app, but "max" not available in the vscode extension picker.
I tried to restart vscode and enable/disable the extension in vscode, no success

[jvalerodev]

@akitin-work you have to click on Codex settings button and then Configuration tab. There you are going to see the available reasoning efforts with "Max" disabled. I thought it wasn't available in VSCode too, but there it is. (Sorry for my English btw).

[wkoo1]

I am experiencing the same issue.
Note that the problem described here concerns the Codex plugin within the VS Code editor, not the Codex client itself.

[jvalerodev]

@wkoo1 I know. The screenshot that I posted is from the VSCode extension. Try to do what I said.

[wkoo1]

@jvalerodev Thank you so much! I tried it, and it really works.

[akitin-work]

@jvalerodev
Thanks — I already have that setting enabled. All six reasoning efforts, including Max, are selected in Codex Settings. However, Max still does not appear in the VS Code extension's reasoning-effort picker. The issue is not that Max is disabled in the settings, but that the VS Code extension does not expose it in the picker.

[aaadashuaige]

You need to open the Codex window in VS Code instead of the Codex desktop app, then go to the settings in the top right corner of the window.

[Firstmeridian]

I don't have reasoning effort settings in the Codex plugin.

[aaadashuaige]

I don't have reasoning effort settings in the Codex plugin.

You can open config.toml and try changing the model and other settings to the following. @Firstmeridian

model = "gpt-5.6-luna"
model_reasoning_effort = "max"
plan_mode_reasoning_effort = "max"
service_tier = "default"
personality = "pragmatic"
sandbox_mode = "workspace-write"
approval_policy = "on-request"