I really like the 3.5 series of models. Every time I compared them against the previous version, I always preferred the new ones. The small models don’t read like 4o anymore, but they do overthink. I disabled reasoning for my use cases.