Fast mode is selected by deafult or switched on request (cloud agents)

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Today, after the update, I noticed that few of my cloud agents was running cursor-grok 4.6 with fast mode. I always have fast mode disabled. Today when I first discovered it, I made sure that next chats for sure have it disabled and they were. Something switches it either on request or next prompts to the chat.

Steps to Reproduce

I don’t have concrete steps to reproduce

Expected Behavior

Fast mode selection on first prompt is respected and not switched mid session.

Operating System

MacOS

Version Information

Version: 3.17.8
VS Code Extension API: 1.128.0
Commit: 2fdd31c9f33f7fbe501f2d57772dc5bf64b63620
Date: 2026-08-20T02:18:12.724Z
Layout: Agent Window
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.291
OS: Darwin arm64 23.3.0

For AI issues: which model did you use?

cursor grok 4.6

Does this stop you from using Cursor

No - Cursor works, but with this issue

+1 on this

Also the case in automation where fast is really not needed but picked for the run regardless of the saved choice on the automation

Hey @Marcin_Miler
You’ve already got Fast off and it’s still flipping on, so this isn’t on your side - the model choice isn’t sticking the way it should, and it’s an issue we’re tracking.

Worth trying: pin it at the model level (picker → hover Cursor Grok 4.6 → Edit → Fast off), which new agents inherit better than a per-chat toggle, and update to the latest version from cursor.com since there’ve been changes in this area recently. If an agent still switches after that, drop a link to it and the rough time and we’ll dig in. Same goes for the automations case - thanks!

dealing with this too

+1

I’m trying really hard to save on usage and this is not making it easy!

Yeah guys I have been speaking on this with Brett from support and so far nothing. The fact that we cannot control what models are being used regardless of auto routing or whatever optimization is being used is ridiculous.

I have suspicion that this switch to fast mode happens whenever cursor servers are overloaded and grok models are slower, the same happended today multiple times. And as I said before I have everywhere this option unselected, still sometimes it’s automatically switched on to fast mode :stuck_out_tongue: Both for main chat model and explore subagents (which are configured to cursor-grok 4.6 high)

This is screwing us over too. Cloud agent sessions will randomly switch back to having fast enabled. Really frustrating when it burns through 2-3x as much token usage allowance.

Same bug, but from Grok Bot launching a Cloud Agent.

Grok bot passes an explicit non-fast override on every launch:

  • model: grok-4.6
  • model_params: { “fast”: “false” }

(API shape: model.id = grok-4.6, model.params = [{ id: “fast”, value: “false” }])

I also pinned Fast off on Cursor Grok 4.6 (picker → hover → Edit) and set the Cloud Agents dashboard default to non-fast.

The agent still starts as Grok 4.6 Fast in the UI. Reproduced three times today, last one 10:44 UTC (cancelled as soon as Fast showed up). fast=false is a documented catalog param and it is being ignored on create.

I’m having the same problem with Grok Bot spinning up Cursor agents. Its burning through my usage allowance at an insane rate.

I tried setting the default model in Cursor cloud to Grok 4.6 high. The bots have tried explicitly passing false for fast mode. Nothing works.