As Subagent it works with my Auto+Composer 
Pro, API 100%, Auto+Composer 52%, Grok blocked
Hi all, we are pushing a fix shortly that will correct this so that Cursor Grok 4.3 pulls from the Auto+Composer pool. @Fernando-Fernandes @Jonathan_Ilagan @jiayu
Hahaha, so you’re suggesting I use your vibe-coding product? That’s hilarious… 
We are putting out a fix for this very shortly, thank you for your patience. @komehdaht Cursor Grok will pull from the Auto + Composer Pool once resolved.
It’s not mine:
~26$ for Grok 4.5 High Subagents + Composer 2.5 Verifiers and ~4.3$ for GPT-5.5 Xhigh leader and GPT-5.4 High final Verifier
Because of what happened between Grok 4.0 and Grok 4.5, I have a strong distrust of Grok, but I hope that GPT-5.4 can be retired and replaced with Grok, and GPT-5.5 replaced with GPT-5.6 Sol (Ultra?) tomorrow.
Hi all, I’m back! back with a quick update: the first-party usage attribution issue for Cursor Grok is fixed. Selecting Grok 4.5 in the model picker should now draw from the recently renamed Auto + Composer pool (not the API usage pool). Note that the Auto+Composer Pool, as part of this change, as been renamed to the 1st Party Model pool.
If you were blocked earlier because API usage was exhausted while the 1st Party Model pool still had headroom, try again, and it should work as expected.
@Fernando-Fernandes @Jonathan_Ilagan @jiayu @komehdaht @luthfihizb
Thanks for the reports and for your patience while we pushed the fix.
Why does grok 4.5 require max mode but Opus 4.6 doesn’t even though it’s more expensive? You guys want us using your competitor instead? I’m sure Elon would love that
Cursor Grok 4.5 requires Max Mode for legacy request-based plans.
But opus 4.6 doesn’t and it costs more, that doesn’t really make any sense?
So far so good. I’m using it right now for website competitive analysis, capturing content on each, analyzing for conversion etc. It’s opening each website in browser and scrolling, capturing visual details competently. It also built a whole website mockup using my content spec super quick.
However, had one hiccup – it tried to bulk splice in a bunch of content to our markdown doc with Python, and that caused the .md preview to break. Even restarting cursor didn’t help. Had it try to fix the .md file manually to no avail so we reverted to before that python splice, which fixed it. I asked it what the issue was:
“So: content was fine; the bulk splice + stuck preview was the likely culprit. For big markdown inserts in this repo, prefer
StrReplace/ smaller edits over rewriting the whole file in Python.”
Just turn off legacy pricing and move on
What about privacy. I haven’t found anything on except this:
https://forum.cursor.com/t/grok-code-fast-1-data-retention-training-policy-free-period/144249/4
It’s important and one reason I use Cursor. Does that still apply since the purchase?
I paid for an annual plan with legacy pricing, why would I switch to a pricing structure that’s actually worse? Plus keeping opus 4.6 out of max mode but forcing max mode on grok 4.5, makes 0 logical / business sense, they are paying a premium to let legacy users use their competitor, imagine spending 60 billion dollars and you implement a business policy that makes original users use your competitor and your business pays more for it, but locks down your very own product behind a pay wall for those same legacy users
that’s one of the most stupid business policies I’ve ever come across honestly
What is the legality behind this? When we bought the subscription, did not you advertise the plan as frontier intelligence with 500 fast request and unlimited slow requests or not? Now every new model, not only frontier models are hidden behind MAX ONLY non-sense.
How come Cursor betrays their early users and try to find a loophole to pasify them? You applied this to external models, but applying it to Cursor models is extremely concerning…
Community is getting mad, I am sure law suits will be filed as this is clear breach of the contract.
the logic makes 0 business sense either
“Here use opus 4.6 and we’ll subsidize the costs! but want to use our very own product? you’ll need to pay full price + a premium for that”
The fact they’d rather people use their direct competitor is insane too, imagine spending 60 billion and the company you just bought implements a policy that makes your own users use your competitor over your own product, it’s pure stupidity
if you told this to any decent economics professor/business person with an IQ over 60, they’d have a stroke and ask if you’re high
Would love to try but its MAX mode only and I can’t upgrade to Team Premium as per Cursor Support. This is so messed up.
Hi @julian_miller, Thanks for the question. This model is treated the same as other Cursor models; there is no change to your privacy settings. Customer Data for customers in Privacy Mode is not used for training, and that applies to Cursor Grok 4.5 the same as any other model. For more information, see Cursor · Data Use & Privacy Overview