Moondream 2

Moondream 2: The tiny, open-source Vision Language Model (VLM) engineered for efficient, real-time image understanding on edge devices.

This is Moondream 2: a highly efficient, open-source VLM available in 2B and 0.5B parameter variants. Built on a foundation of SigLIP and Phi-1.5 weights, the model delivers robust visual intelligence for resource-constrained environments (edge devices, local deployment). It handles core vision tasks—visual question answering (VQA), image captioning, and zero-shot object detection—with speed and accuracy, making it a critical asset for robotics, UI automation, and scalable enterprise analytics.

https://moondream.ai