A vision-language reranker from the Qwen3 family with 8 billion parameters, capable of re-ranking both text and image inputs.