m87-labs/moondream

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

10014 stars · 793 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

Moondream is an open-weight family of small, efficient vision language models (2B to 9B MoE) that perform image captioning, visual question answering, object detection, pointing, counting, and segmentation. It ships with a Python package and the Photon local inference engine for NVIDIA GPUs and Apple Silicon, plus a hosted cloud API and fine-tuning service.

Use cases

• caption images with a small local model
• run visual question answering on edge devices
• detect and locate objects in images with bounding boxes
• segment objects in images without heavy GPU hardware
• fine-tune a vision model on my own labeled images
• run a VLM on a Jetson or Apple Silicon Mac
• count objects in photos
• understand documents and charts with a vision model

When to choose

• you need efficient image understanding on constrained hardware or at the edge
• you want open weights with commercial use allowed for local inference
• you need vision skills like detection, pointing, and segmentation out of the box
• you want low inference cost compared to frontier VLMs

When to avoid

• you need frontier-level general reasoning or long-document understanding beyond its benchmarks
• you plan to host the model weights as a third-party cloud service, which the license restricts
• you need text-only LLM capabilities without vision

Facets

library · maturity active

machine-learning computer-vision llm-inference image-processing sdk computer-vision artificial-intelligence machine-learning image-processing python windows cloud self-hosted vision-language-model vlm image-captioning visual-question-answering object-detection image-segmentation edge-ai mixture-of-experts open-weights fine-tuning linux macos gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="m87-labs/moondream")