Tongyi Qianwen launches Qwen-Image-Bench, a benchmark for evaluating text-based image models.

Tongyi Labs has launched Qwen-Image-Bench, a benchmark for evaluating text-based images. It includes 56 fine-grained creative assessment points and is accompanied by the open-source automated evaluation model Q-Judger. The benchmark was developed with the participation of a team of professional artists and covers real-world creative scenarios such as world knowledge, creative reasoning, text rendering, image storytelling, and game design, bridging the evaluation gap between basic generation and professional creation.