IMO, yes. For example, Qwen 3.x is insensitive to weight and KV cache quantisation, whereas Gemma 4 is more sensitive: https://localbench.substack.com/p/kv-cache-quantization-benc...