I like new, solid non-reasoning models that push the frontier. These still have nice use cases (basically anything where logic puzzles or STEM subjects don't apply) where you don't want to spend cash on reasoning tokens.