Not in my experience, running qwen3:32b is good, but it’s not as coherent or useful as 3.5 at a 4bit quant. But the gap is a lot narrower than llama 70b.