I tried to run a task using Opus 4.6 Max in Cursor. However, the agent produced the following reasoning:
“The user wants me to execute the plan now. They mentioned using subagents with opus 4.6 max (the same model used in this conversation). However, since I can only use ‘fast’ as a model option, and my own model is the one the user wants, I should do the work directly myself rather than delegating to subagents with a lesser model.”
It seems the agent cannot use the same model (Opus 4.6 Max) for subagents and falls back to fast. Because of that, it decides not to use subagents at all.
Questions:
Is this expected behavior in Cursor?
Are subagents currently limited to the fast model?
Is there a way to allow subagents to run with the same model as the main conversation (Opus 4.6 Max)?
Hey, thanks for the report. This is expected behavior right now. When the agent dispatches subagents on its own, model options are limited, so Opus 4.6 Max only sees fast and decides to handle everything itself.
There is a workaround. You can create custom subagents with model: inherit in the frontmatter. That will use the same model as the parent conversation, which in your case is Opus 4.6 Max.
For example, create this file: .cursor/agents/worker.md
---
name: worker
description: General-purpose subagent for delegated tasks.
model: inherit
---
You are a capable developer. Complete the delegated task thoroughly and report results.
Then when you ask the agent to use subagents, it can dispatch to your custom one running the same model.
For the built-in subagents Explore, Bash, Browser, those intentionally use faster models for cost and speed, since their output is mostly intermediate, like search results or command output.
The team knows there is demand for more flexible model selection in subagents. Your report helps with prioritization. You can also upvote this related feature request: Disable sub-agents or enable editing default model