>>105882039
it struggles to be better than most 30b dense models with frequent logical mistakes that really do feel like a 17b model. Also, its very censored and no one has mad a sloptune of it that I know of.

17b active params is just not enough, and scout is a greedy product—meta clearly wanted something very cheap to run as their free ai for the masses. R1 having 37b active is probably why its decent.