Repository navigation
Subagents panel no longer shows each agent’s model or reasoning effort #32283
Description
What version of the Codex App are you using (From “About Codex” dialog)?
26.707.3748.0
What subscription do you have?
ChatGPT Pro (20x)
What platform is your computer?
Microsoft Windows NT 10.0.26200.0 x64
What issue are you seeing?
The Codex Desktop Subagents panel shows each agent’s role, status, duration, and result summary, but it no longer shows the model or reasoning effort the agent actually used.
This was visible in the earlier GPT-5.5-era experience. It matters now because Codex supports role-specific model configuration and different models have separate usage limits.
For example, my parent runs on GPT-5.6 Sol High, while my custom roles map worker to GPT-5.6 Terra Medium and explorer to GPT-5.3-Codex-Spark Low. The panel only says “Worker,” so I cannot confirm whether the configured model was applied, inherited from the parent, or replaced by a fallback.
The subagent completed successfully; this is a visibility/observability regression, not a spawning failure.
What steps can reproduce the bug?
- Add a custom agent configuration such as:
# ~/.codex/agents/worker.toml
name = "worker"
description = "Focused implementation worker."
model = "gpt-5.6-terra"
model_reasoning_effort = "medium"- Start a Codex Desktop task using a different parent model.
- Ask Codex to spawn a
workersubagent. - Open the Subagents panel after the worker starts or completes.
- Observe that the role and summary are shown, but the resolved model and reasoning effort are not.
Session ID: 019f4b76-7459-7850-a975-4ebae680c74c
What is the expected behavior?
Each active and completed subagent should show its resolved model and reasoning effort, for example:
Worker · GPT-5.6 Terra · Medium
Ideally the detail view would also indicate whether those values came from the role configuration, were inherited from the parent, or were selected as a fallback. Restoring the previous model badge would already solve the main problem.
Additional information
This is especially important for Pro users because GPT-5.3-Codex-Spark has a separate usage allowance. Without model visibility, users cannot verify quota-aware routing or diagnose unexpected usage.
Related but distinct: #14965 reports an incorrect subagent reasoning label in the older UI; this report is about the current Desktop panel omitting the model and reasoning label entirely. #27624 concerns configuration being ignored; in this case the configured routing appears to work, but the UI cannot confirm it.
Earlier Codex Desktop UI: the model badge was visible. Public example from #14866.
Current Codex Desktop UI: the completed Worker entry shows its duration and summary, but no model or reasoning-effort information.
Activity
github-actions commentedon Jul 10, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
dsfaccini commentedon Jul 22, 2026
This comment was posted by Codex Desktop using gpt-5.6-sol on behalf of David.\n\nWe're seeing this as well and care about having the model and reasoning-effort labels restored.\n\nWhen an orchestrator routes different subagents to different models or effort levels, the role name alone doesn't let us verify which configuration actually ran. Showing the resolved model and effort for both active and completed subagents is important for trust, debugging, and usage management. If feasible, indicating whether those values were explicitly configured, inherited, or selected as a fallback would be especially useful.
praneeth-red commentedon Aug 17, 2026
Please show the exact resolved model and reasoning effort beside every sub-agent—for example, “Explorer · GPT-5.6 Terra · Medium.” It should also indicate whether the model was inherited, explicitly selected, or used as a fallback. This appears related to openai/codex issue #32283.
MichaelYochpaz commentedon Sep 4, 2026
I'd also love to be able to see what model / subagent type the initiated subagents are.
It's quite important for inspecting workflows and trying to optimize them (e.g., knowing whether the agent initiate Luna, Sol or Astra subagents).
imCinq commentedon Sep 6, 2026
I'm having the same issue as well — the subagent model/reasoning information no longer shows for me.
hanjie-chen commentedon Sep 12, 2026
+1. This matters a lot for quota-aware routing.
I want to keep the parent agent on a stronger model while routing simpler subagent work to cheaper models with lower reasoning effort. Without showing the effective runtime model and reasoning effort in the Subagents panel, I cannot verify whether the intended routing actually happened or whether the child inherited/fell back to another configuration.
Atroci commentedon Sep 29, 2026
+1, and for routing it helps to show the requested and effective values separately, with a reason when they differ:
A panel that shows only the effective model can't tell "role config ignored" from "fallback after an error", and those need different fixes. Without both values it's hard to confirm that a quota-aware setup is actually saving anything.