Comment by XCSme

My concern is that reasoning could involve some sequential steps that instant models don't.

Not sure if modern models "think" only by outputting <thinking> blocks, or there is a more complex mechanism at play.