Comment by WarmWash

I haven't used 3.5 at all yet, but previous Gemini (and Gemma models) are by far the most token light per task than any other model.

Cost per task is a more productive measure, but obviously a more difficult one to benchmark.