when do you switch from a fast cheap model to a heavy reasoning model mid-session?

by•

for standard UI components and basic CRUD endpoints, lighter fast models feel instant and keep my momentum going without waiting thirty seconds per reply.

​the moment a tricky async bug or state loop shows up, those same fast models start going in circles until i switch over to a heavier reasoning model.

​do you stick to one model for the whole build or swap them depending on the task?

17 views

Add a comment

Replies

Best

Switching happens the second an edit requires importing from another file. A fast model lost context mid-task and silently rebuilt an existing helper two files away under a new name, leaving duplicate logic that went unnoticed until code review.

I almost never swap mid session. Since 3 Nov only 6 of my sessions used more than one model, and every other one stayed on the model it started with.

The 6 that switched all had Opus in them, paired with Sonnet or Fable. When you switch for a stuck bug, does the heavier model get the whole failed history, or do you start it fresh with a summary?

For me the switch isn't really about speed, it's about who gets to judge. I let one model do the work, and whenever a model makes a judgment call, a different model checks it. Same thing when one gets stuck in a loop. A fresh model that didn't write the code is usually what breaks the circle.