I am having the same thing. i sometimes notice it switch over just in the middle of a sentence, othertimes it changes after i send the message. I thought I was going insane.
This is a known bug: on Grok 4.5 the per-model Fast toggle doesn’t stick, so it can re-enable after a prompt or new chat and your requests run the higher-cost Fast variant (cursor-grok-4.5-high-fast) even though you switched it off. I’ve logged your report with the team. No fix has shipped yet and I can’t share an ETA.
Workaround until it’s fixed: toggle Fast off right before you submit, and confirm the picker label reads just Cursor Grok 4.5 High with no “Fast”. You may need to redo this on each new chat.
If you’d like us to review the Fast charges that came through while it was mis-toggling, email [email protected] and we’ll take a look.
Hi - I was actually doing it the way you suggest, but a few seconds after the prompt gets run, it switches back to fast. And unfortunately, switching it back after doesn’t stop the token usage under the fast variant which I assume is charged at a noticeably higher rate. Perhaps equalize the billing short-term until this is resolved?
This issue has definitely not been fixed in the latest version on Windows.
Using “Cursor Grok 4.5 High” in any of IDE, Agent Window, or Automations / Cloud Agents always gets labelled as cursor-grok-4.5-high-fast in the Usage report. See screenshot below. I have been using the non-fast version quite a bit in the past few days and it’s not showing on the model usage list. Unless if it’s being grouped into “Other”
Hey @C_ST and @Kris_Gunnars - one part should be sorted now: sub-agent, background, and Automations / Cloud Agent runs no longer bill the Fast variant when Fast is off. That’s a server-side change, so nothing to update on your end.
The Fast toggle still flipping back on in the model picker (IDE / Agent Window) is a separate issue we’re tracking - while that happens, requests where the picker shows “Fast” can still run Fast, so confirm it reads just Cursor Grok 4.5 High before you submit.
For any Fast charges that came through while this was misbehaving, email [email protected] and we’ll review them.
Mohit - thanks for the update.
One other thing I’ve noticed.. new chat window seem to often default to Grok now, latest release. I can’t reproduce it reliably, but I’ve noticed a few times I launched a prompt expecting Auto and Grok was pre-chosen.
You can turn that off: Settings → Agents → Default Model → Last Used, and new chats will keep your last-used model (like Auto) instead of suggesting Grok. Worth a quick glance at the model pill before submitting, too.
This isn’t something you misconfigured - it’s not intended behavior and it’s an issue we’re tracking. For any Grok charges that came through while you were expecting Auto, email [email protected] and we’ll take a look.
