> qwen models (qwen3.6-35b-a3b and earlier 3.5) are WAY better with vision than gemma4-26b-a4b

Can you give an example? And/or is there a benchmark specifically for this?