Xiaomi Releases Open-Source Xiaomi-Lens-0 Model with 4 Billion Parameters

Xiaomi has announced the release of Xiaomi-Lens-0, a multimodal model with 4 billion parameters that can pinpoint the exact image region while answering. The model is available on Hugging Face under SeerRay-Lab, with a technical report on arXiv, showcasing the company's commitment to AI development. Built on Qwen3.5-4B, the model incorporates a lightweight pixel-mask decoder. Xiaomi-Lens-0 supports various tasks such as referring expression comprehension and segmentation, making it suitable for robotics experiments and visual assistants in phones. Xiaomi also released the LensBench segmentation benchmark for evaluation.