Cursor's /review-bugbot subagent always bills as Composer 2.5 Fast

If you run /review-bugbot or /review-security in Cursor to save money on standard Composer 2.5, the review subagent still runs, and bills, on Composer 2.5 Fast. Cursor staff member kevinn confirmed at 2:16 PM ET on Oct 16 that the subagent "is currently set to run on Composer 2.5 Fast" and "doesn't follow the parent chat's model or the model value it's launched with, including inherit." There's no setting for it yet, so use one of the two workarounds below.

What changed#

Nothing new shipped; this confirms how it already works. Igor_Tomalski reported at 7:41 AM ET that his usage log showed a composer-2.5-fast row for a review he started on plain Composer 2.5, charged about 39 cents, "instead of ca. 9 if subagent was working as composer 2.5." kevinn said it "isn't a display issue." The parent chat stayed on your model; only the subagent part billed as Fast. An older review in July didn't launch a subagent, so it ran on the chat's model, kevinn said.

Why it matters#

Fast costs much more. Cursor's pricing lists Composer 2.5 at $0.50 per million input tokens and $2.50 per million output tokens, and Composer 2.5 Fast at $3 and $15, six times as much. Cache reads are $0.20 versus $0.50, so the blended gap is smaller; the reporter's run came to about 4x. Those rates apply to usage beyond your weekly pool. Our read: a review reads your branch diff, so the gap grows with every review you run.

How to keep it on Composer 2.5#

kevinn gave two ways:

1. Review in the main chat. Pick Composer 2.5 with Fast off, then ask: Review my branch changes for issues yourself. Do not launch any subagents.
2. Use your own review subagent. Create .cursor/agents/my-review.md in your project:

---
name: my-review
description: Reviews local branch changes for issues. Use when I ask for a code review.
model: composer-2.5[fast=false]
readonly: true
---

(your review instructions here)

Then ask the agent to "use the my-review subagent to review my branch changes." The [fast=false] part picks the standard model, per Cursor's subagent docs.

Good to know#

• kevinn said the same applies to /review-security.
• After your first run, check your usage details. The row should say composer-2.5, not composer-2.5-fast.
• This thread is about the /review-bugbot and /review-security commands you run in the editor. Staff did not say anything about Bugbot reviews on pull requests.

What to watch#

kevinn passed the feedback to the team and said he'll post in the thread if the subagent's model changes. A model setting for the review subagent would be the sign.