Subagent `resume` runs an incorrect model

Describe the Bug

When calling a Subagent with resume, the Subagent inherits the model from the Agent, not from the resumable Subagent. Reasoning effort is not inherited.

Steps to Reproduce

with not the composer-2-fast model:

Hello.
1. Request a connection test and answer "OK" from the model composer-2-fast
2. Then, resume to the same model asking for "hello world."

Screenshots / Screen Recordings

Operating System

Windows 10/11

Version Information

Version: 3.4.1 (system setup)
VSCode Version: 1.105.1
Commit: 2a298dd06944a9b9ea541d28225b779fcbcc6200
Date: 2026-05-08T16:05:07.818Z
Layout: editor
Build Type: Stable
Release Track: Nightly
Electron: 39.8.1
Chromium: 142.0.7444.265
Node.js: 22.22.1
V8: 14.2.231.22-electron.0
OS: Windows_NT x64 10.0.22631

For AI issues: add Request ID with privacy disabled

8b99cc0c-030e-4438-845e-f350fbce28fe
ba658273-3c58-4b02-8a2c-9c875c4a9e35

Additional Information

Here are the exact payloads for the four Subagent calls.

1) First run

{
  "description": "Connection test with composer-2-fast",
  "prompt": "Connection test only. Reply with exactly: OK",
  "model": "composer-2-fast",
  "subagent_type": "generalPurpose",
  "readonly": true,
  "run_in_background": false
}

2) Resume of the first run

{
  "description": "Follow-up hello world prompt",
  "prompt": "hello world.",
  "resume": "afbb4674-a581-4066-a60c-afba77445a0c",
  "subagent_type": "generalPurpose",
  "readonly": true,
  "run_in_background": false
}

3) Second run

{
  "description": "New connection test on composer-2-fast",
  "prompt": "Connection test only. Reply with exactly: OK",
  "model": "composer-2-fast",
  "subagent_type": "generalPurpose",
  "readonly": true,
  "run_in_background": false
}

4) Resume of the second run

{
  "description": "Resume with hello world prompt",
  "prompt": "hello world.",
  "resume": "90e0aaba-34a4-403a-be1f-ba746ac5c811",
  "subagent_type": "generalPurpose",
  "readonly": true,
  "run_in_background": false
}

If you want, I can also give you the same payloads as a compact one-line JSON list.

Does this stop you from using Cursor

Yes - Cursor is unusable

I would be extremely grateful if I could be notified when the bug has been fixed, as I would now have to add a note to the instructions prohibiting the use of resume.

This is a confirmed bug. When a subagent is resumed via resume, the model from the original run is not preserved — it falls back to the parent agent’s model instead. Your usage log confirms gpt-5.4-mini-medium was actually used on the resumed calls instead of composer-2-fast.

This is separate from the display bug we discussed in your earlier thread — here, the wrong model is actually being used, not just displayed incorrectly.

I’ve escalated this to our engineering team. We’ll update this thread when a fix ships. In the meantime, spawning a fresh subagent with an explicit model parameter instead of using resume is the right workaround.

Solved in Nightly 3.5.3

Broken again in 3.5.11

Thanks for reporting back. This is a regression — the fix that shipped in 3.5.3 was inadvertently dropped during a subsequent refactor. I’ve flagged this on the tracking ticket so the team is aware. A re-fix should follow.

v3.6.12
Sometimes it works correctly, sometimes it doesn’t.

In v3.7.6 with GPT-5.5 as Agent and Composer 2.5 as Subagent works good.