Question about Cursor's Tab Completion Model

Discussion
1

Hello Cursor community,

I have a question about the model used for Cursor’s Tab completion feature (the code suggestion feature in the editor). According to the documentation, this feature is powered by a “custom model,” but I couldn’t find specific details about which model it actually uses.

I noticed that there’s a “cursor-small” model listed in the available models. Is this the model that powers the Tab completion feature? Or is it a different model?

Additionally, the documentation mentions that Tab attempts to make suggestions with every keystroke and cursor movement. Does this mean it’s sending requests to a cloud-based model each time? If so, does this count toward usage limits?

Any clarification would be greatly appreciated.

Thank you!

1 Like
2

I’d also want to know which “custom” model Cursor tab is using.

As for tab usage limits, their docs mention that Pro and Business users get unlimited tab completions.

cursor-tab
cursor-tab710×265 5.82 KB

3
  1. They use the so called Fusion model to support the Tab Completion，as described in their blog:

this model is a specific builtin model dedicated for the tab auto completion, so you cannot see it in the model list.

  1. Yes, every auto completion request is sent to their could hosts, which you can observe the requests via the Output => Cursor Tab

  2. No, these auto completions or tab completion are not included in the usage limits, which means if you are a pro/business user, you have unlimited auto/tab completion.

3 Likes
4

Thank you so much for your detailed response about the Tab completion functionality in Cursor!

Your clarification about the Fusion model has been extremely helpful. It’s fascinating to learn that Cursor uses a dedicated model specifically built for tab auto-completion that doesn’t appear in the regular model list.

I particularly appreciate you confirming that every auto-completion request is indeed sent to cloud hosts (which can be observed in Output => Cursor Tab) and that these completions don’t count toward usage limits for Pro/Business users.

The blog post about the Fusion model was also very informative - understanding how it not only predicts edits but also suggests cursor jumps gives me a much better picture of the technology behind this impressive feature.

This information has completely resolved my questions and helped me better understand how to make the most of Cursor’s capabilities.

Thanks again for taking the time to provide such a comprehensive answer!

Best regards