A large multimodal model built on Qwen3's vision-language foundation, this variant has been 'abliterated' — a community fine-tuning technique that removes refusal behaviors, making it more permissive in responding to a wider range of requests. It handles both text and image inputs, combining visual understanding with language generation. The trade-off is that safety guardrails present in the original model have been deliberately stripped out.