Qwen 3.8 still has a problem, but not in the way you're thinking.
Image-Text-to-Text model with 24B parameters, built with vllm. Download, explore files and run it on Hugging Face.