Inquiry: AI Model Utilization in Cursor's Different Modes

General
1

Hello everyone,

I’ve been quite intrigued by how Cursor employs different AI models for various functionalities such as Tab, Diff, and Chat.

From what I’ve gathered, the Chat feature defaults to using the Claude 3.5 Sonnet model, likely due to its prowess in language handling and code suggestion accuracy. For the Tab scenario, to enhance speed, I speculate that Cursor might employ specially optimized models that can quickly provide code completion suggestions as users type. As for Diff, given its requirement to handle detailed code comparisons and format alignment, it might utilize models specifically fine-tuned for these tasks.

I’m seeking confirmation on whether this information is accurate and if there are any official resources or community insights that could offer a deeper understanding.

2

Chat uses the model you’ve selected from the dropdown. Tab and Apply use custom models.

3

Follow-up question: I have access to an Azure OpenAI deployment. Am I right to assume that if I wire that up using the Azure API Key feature, only the Chat will be affected, not the Tab feature? More generally, do the settings in the Models section of Cursor Settings affect only the Chat, and Cursor Tab will always use Cursor proprietary models?