Share your thoughts on the new reasoning model

The new reasoning model has been the default for a week. How is it working for you?

For me: much better at multi-file refactors, noticeably slower for small edits, so I switch to the fast model for those.

Agree on both. I set the fast model as default and only pick the reasoning one for planning.

It is the first model that understood our pandas pipeline without me pasting the schema.

It writes much better tests. Fewer mocks, more real assertions.

Slower, but fewer back-and-forths overall, so net faster for me.

Sometimes it overthinks a one-line fix and proposes a refactor. A "just do the minimum" toggle would help.

A "just do the minimum" toggle would help.

Thanks all, this is really useful. The minimum-change idea has come up a lot, we are looking at it.