MiMo-VL-7B

MiMo-VL-7B is Xiaomi's open-weight vision-language model, released in June 2025 by the team led by former DeepSeek researcher Luo Fuli. It pairs a native-resolution vision transformer with an MLP projector and the MiMo-7B language model, trained with multi-stage pre-training and mixed reinforcement learning to reason across text, images, video and UI elements.

Xiaomi's vendor-reported figures put it above 70 on the MMMU multimodal benchmark, with 94.4 on ChartQA, 92.5 on ScreenSpot-v2 for GUI tasks and 70.8 on VideoMME, results the company frames as competitive with models ten times its size. Weights are openly available for researchers and developers. As with any vendor-reported benchmark, treat the headline numbers as the company's own claims and check the model card for license terms.