Which model in Windsurf for a big refactor vs small fixes? I'm burning credits

What works for me:

• Small, well-described edits (copy change, CSS, add a prop, rename): the fast in-house model. It's cheap and honestly just as good when the task fits in one file.
• Planning the refactor: the strongest model, once, in a chat where it may not write code. Ask for a step-by-step plan with file names.
• Executing the plan: mid-tier model, one step per message, run your tests between steps.

The expensive part is not the model choice, it's letting a big model loop on a vague task for 25 tool calls.

I'd push back a little on mid-tier for the refactor execution. Splitting a 900-line file is where the cheaper models tend to "helpfully" rewrite logic while moving it, and you only notice in production.

For moves specifically I use the strong model but with a tight prompt: "move functions X, Y, Z verbatim to api/users.ts, update imports, change nothing else". That's short output, so it's not that expensive. Credits go on long outputs and long loops, not on the model name alone.

What nobody tells you: the cheaper models are fine for more than people think, as long as the task is small. Most of my credit waste was vague prompts, not model choice. 'Make the dashboard better' costs 10x more than 'move the filter bar above the table'.