Moondream 2

Moondream 2 is a tiny vision-language model by vikhyatk designed to run efficiently everywhere, from Raspberry Pis to AI PCs. At 1.86 billion parameters, combining SigLIP vision weights with a Phi-1.5-based language model, it needs under 6GB of memory and runs on CPU as well as GPU.

It supports image captioning, visual question answering, object detection, and point grounding through a simple Python API or Hugging Face Transformers. Released under Apache 2.0, it became a popular choice for real-time camera applications, accessibility tools, and on-device visual assistants where larger VLMs are impractical.