I like to use the fast model. I don’t think I will hit the limit of my plan based on projected usage, so I am happy to pay more for the extra speed.
Between [gpt-5-fast] and [gpt-5-high-fast] I honestly haven’t noticed any difference at all. The work I’m doing isn’t particularly hard for Cursor to complete. I know I’m in good hands with these models.
I think TypeScript works great with Cursor. The LSP and linter integrate nicely for picking up errors, and there are patterns everywhere for the LLM to follow.
I do prefer Angular over React and frontend isn’t my strong area, but I chose React because there is maybe so much more training data for it and I find with using a opinionated UI library like Material UI, the code is always spot on. I don’t think I’ve manually written a line of frontend code on this project, which is wild to say! That is the benefits of working across a full TypeScript monorepo as I can give context from the backend schema and the frontend can follow very nicely.
Here is a post from a while back explaining alot more which you may find useful: