I tried Qwen3.5 9b & 4b versions for the last couple of days and my experience has been really bad. Qwen3.5 regardless the size are slow and seems like overthinking too much. I’m getting better results with ministral-3.
I’ve had decent results with Qwen3.5-122B-A10B-AWQ-4bit. But I’m planning to try the latest unsloth version to see how it performs. There are times when I think I’m talking to a child especially when I come off of using Opus.