Moondream2
A tiny vision-language model that runs almost anywhere.
About
Moondream2 is a 1.86-billion-parameter vision-language model built to run on modest hardware, including CPUs and edge devices. It handles captioning, visual question answering, object detection, and point localization from natural-language queries, which makes it handy for document OCR and lightweight image understanding. It is Apache-2.0 licensed, genuinely open for commercial use, and the weights are updated regularly on Hugging Face. It is not a frontier model, so it trails larger VLMs on hard reasoning, and its MPS support on Apple Silicon has had issues.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
DocVQA 79.3, TextVQA 76.3, OCRBench 61.2 in a third-party A10 benchmark run; VQAv2 79.0% vs GPT-4o 77.2% reported by Roboflow (2024).
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Widely deployed tiny VLM; Roboflow guide and broad community usage; Weights updated regularly (latest revision 2025-06-21); Moondream 3 preview now exists; Apache-2.0, confirmed on the HF card; commercial use allowed; Benchmark figures cited are from community/third-party runs, not a single official eval sheet. License: Apache-2.0 per the Hugging Face model card..
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.0/10
- Maintenance · recency, releases, issue load
- 7.0/10
- Openness · license + docs
- 9.0/10
- Manual review · sanity read + red-flag check
- 7.0/10
- · Widely deployed tiny VLM; Roboflow guide and broad community usage
- · Weights updated regularly (latest revision 2025-06-21); Moondream 3 preview now exists
- · Apache-2.0, confirmed on the HF card; commercial use allowed
- · Benchmark figures cited are from community/third-party runs, not a single official eval sheet