Cloud Agents can't spin up "fast" model subagents correctly

Support Bug Reports
, , , ,
1

Where does the bug appear (feature/product)?

Background Agent (GitHub, Slack, Web, Linear)

Describe the Bug

I have a daily scheduled automation that runs Opus 4.6 to post a daily brief for me in Slack. The instructions have it spin up subagents using the “fast” model selection to gather info in their domains, since Composer 2 is good enough, but I still prefer having Opus 4.6 as the orchestrator. It’s working great overall, but I noticed today that the subagents appear to be using Opus 4.6. I can’t see the actual tool calls, but the parent agent insists that it used “model=fast” on every subagent, so that would indicate that the subagent infra isn’t respecting the model choice, or maybe the parent agent is getting the tool call wrong? Here’s a recent example: https://cursor.com/agents/bc-ed3f2b9e-744c-4089-9de7-63f1d29e98d6?branch=cursor/daily-brief-slack-post-982a&app=code

I tried having an IDE agent spin up a subagent using fast model, and that appears to work (ca852533-c71c-472e-bfe1-a7e857a592d1), but another test cloud agent showed the same thing (), so maybe the issue here is specific to cloud agents? (https://cursor.com/agents/bc-40fb0705-47b4-403f-a11d-501f604e3ba8?branch=cursor%2Ffast-model-subagent-health-4b53)

Steps to Reproduce

Fire up an Opus 4.6 cloud agent and tell it to spawn a subagent with “fast” model choice, observe that the subagent also uses Opus 4.6

Expected Behavior

The spawned subagent should use the fast model choice.

Operating System

MacOS

Version Information

Version: 2.6.22
VSCode Version: 1.105.1
Commit: c6285feaba0ad62603f7c22e72f0a170dc8415a0
Date: 2026-03-27T15:59:31.561Z
Build Type: Stable
Release Track: Default
Electron: 39.8.1
Chromium: 142.0.7444.265
Node.js: 22.22.1
V8: 14.2.231.22-electron.0
OS: Darwin arm64 24.6.0

Does this stop you from using Cursor

No - Cursor works, but with this issue

6

Hi @ryanw

I checked our server logs for your test cloud agent, and the “fast” model selection is working correctly. The subagent resolved to Composer 2 on the backend, as expected. The parent agent is passing model=fast correctly, and the infrastructure is respecting that choice.

The issue is that there’s currently no way to confirm which model a subagent used from the cloud agent UI. Since you can’t see the actual tool call details, it’s understandable why this looked off. Your IDE test also confirms things work on that side, and the cloud agent is doing the same thing under the hood.

I understand the lack of visibility here is frustrating, and I’ll pass this feedback along to the team.