Grok 4.5 seems to have a lot of trouble using the AskQuestion tool like Composer 2.5. This happens with prompts triggered from IDE.
Doesn’t matter as much but not as ergonomic.
Maybe problem in your User Rules.
You might try again with my Agent Compass
Did you increase the price for grok or somehting?
It seems to chew through tokens like nothing now and for not very long running tasks either.
It’s been overall great for me. I’ve never used a Grok model in Cursor and amazed at how good it is and the price is outstanding. Its my daily driver…no need to use Opus or 5.6 for planning.
Hi @Calvin_Ng Thanks for the post! This is expected with Grok 4.5 right now. @alirng1
The structured question picker (AskQuestion) is intentionally not available for Grok 4.5, including in Plan mode. When it needs clarification, it asks inline in the chat instead.
If you want the multiple-choice question UI, switch to Composer 2.5 (or another model that still has that tool). Plan mode itself is fine; this is specific to Grok 4.5. AskQuestion may be added back in future models - we’ll let you know!
Glad to hear it, Michael! We appreciate your feedback.
Hi @skaro We have not increased the per-token price for Grok 4.5. It is still $2/$6 per million tokens standard, and $4/$18 for Fast: Grok 4.5 | Cursor Docs
btw, we doubled Cursor Models usage on all individual and team plans last month. Lots of headroom for more coding!
Thank you for the feedback! Best thing you can do in this situation is to send us feedback on the model’s response with the Thumbs Up or Down button.
You can optionally attach more information after selecting a Thumbs-Up or Thumbs down as follows:
This gets the feedback to the right team and directly helps us correlate it with your request.
Yes, that is exactly right!
Hi Charles, we’ve released a fix for this bug today. @des1 @Charles_Wood
A model you’ve turned off in Settings > Models should stay off. Cursor should not re-enable it, select it for a new chat, or switch you to it when another model hits a limit.
That issue is fixed in the latest version. After you update, disabled models stay disabled, and automatic switches no longer pick them. If your current model hits a limit and the suggested fallback is one you’ve turned off, you’ll be asked to choose a different model instead of being switched.
I remember that Grok 4 impressed me with its engineering approach and was the best model between Gemini 2.5 Pro and GPT-5 releases. Best wishes to your team for repeating that success with the upcoming release of Grok-4.6
Grok usage is banned at the company I am working for right now. I cannot change this as it is compnay policy and it took me long enough to get Composer allowed. I do use Grok for personal projects via the same Cursor instance. One thing I have seen is that Cursor is very fond of re-selecting Grok at any opportunity - I ran out of API credits - it selected Grok, I switched to Composer, later I upgraded and Grok was selected again.
Can we get more control of what model to use in these fallback scenarios? I want to keep Grok allowed for manual selection but not auto select. I do not use Auto mode at all - I select my model explicitly.
I’m not a vibe coder so I’m not sure how these models perform in that regard, but for shorter slices of work:
I was not happy with composer 2.
I was happy with composer 2.5.
I am happier with Grok 4.5.
Cursor itself is now a Musk product.
I noticed this morning that Grok is no longer listed when you use it. Other non-xAi models display their names. Grok just displays high/fast. The point being, Cursor is headed towards being a fully SpaceX product with other models being relegated to second-class status.
Thanks Kevin! Can you state the specific version it was fixed in? I am sharing this update with my team.
Well this thread specifically requested feedback 
edit: That feed back feature is Cursor controlled? Because I don’t see a way to trigger it in the code IDE Agent Window. I find Grok consistently over complicating topics and changing code unnecessarily. Having just had another one of these incidents, I chose Feedback from the … Agent menu and do not get this same mechanism.
Overall, I find the feedback survey often comes too early in an Agent conversation and so I have to close it because the result is not yet available. I’d be happy to give both positive and negative feedback.