Opus spawns only Opus subagents, despite different instructions

Support Bug Reports
, ,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Claude Opus models ignore the model parameter on Task subagents — all spawned agents run as Opus regardless of the specified model slug.

Steps to Reproduce

  1. Set chat model to Claude Opus 4.6
  2. Use the Task tool to spawn a subagent with an explicit model parameter, e.g. model: “gpt-5.6-sol-high” or model: “cursor-grok-4.5-high-fast”
  3. Observe the subagent model label in the UI — it shows “Opus 4.6 High” instead of the requested model
  4. This does NOT happen with other parent models (GPT, Composer, Grok) — they correctly respect the model parameter and spawn the requested model

Expected Behavior

Subagents should run on the model specified in the model parameter (e.g. gpt-5.6-sol-high), not inherit the parent’s Opus model.

Operating System

Windows 10/11

Version Information

Version: 3.13.21 (user setup)
VS Code Extension API: 1.128.0
Commit: 55434bd8062ece6fee083b82beed2aee42d253f0
Date: 2026-07-27T03:26:14.573Z
Layout: IDE
Build Type: Stable
Release Track: Default
Electron: 40.10.3
Chromium: 144.0.7559.236
Node.js: 24.15.0
V8: 14.4.258.32-electron.0
xterm.js: 6.1.0-beta.256
OS: Windows_NT x64 10.0.26200

Additional Information

Reproducible in a multi-model ensemble workflow where the orchestrator spawns 3 Task agents with different models. When Opus is the orchestrator, all 3 show as “Opus 4.6 High” in the UI. When Composer is the orchestrator, Sol and Grok run on their correct models. I am using BYOK anthropic api key. Operating system is Windows 11, but i am using WSL.

Does this stop you from using Cursor

No - Cursor works, but with this issue

5

Hey, thanks for the detailed report with the version and steps. This behavior is expected, even if it’s not obvious because of BYOK.

When your own Anthropic API key is active, sub-agents inherit not only the parent model, but also its credentials. That means they run through your key. An Anthropic key can’t run gpt-5.6-sol-high or cursor-grok-4.5-high-fast, so with Opus as the orchestrator all Task sub-agents get pinned to the parent model (Opus). The UI isn’t lying, the sub-agents really ran on Opus.

With Composer, GPT, or Grok as the orchestrator, the session for those models doesn’t go through the Anthropic key, so the model parameter is respected. That’s why you see the difference.

How to run a mixed-model ensemble:

  • Use a non-Anthropic orchestrator (Composer, GPT, or Grok). You already found this.
  • Or disable the Anthropic API key for that session. Then sub-agents will run through Cursor and the model parameter will be applied.

If any workaround doesn’t work, tell me and we’ll dig in.