Moondream
#19 in Multimodale ModelleMoondream · siet September 18, 2025 · 2× · tolest 15. Okt. 2026
25
Momentum
Moondream is a vision-language model (VLM) from M87 Labs, Inc. that analyzes image content and generates natural-language answers (e.g., captioning, visual question answering, object detection). The current version, Moondream 3 (Preview), uses a mixture-of-experts architecture with 9B total parameters, 2B active per token, and is available as open weights and via a cloud API. The product is offered at moondream.ai; according to the announcement, Prime Intellect is acquiring the project.
Momentum-Verloop
20.07.18.10.
Features
| Key Benchmark (%) | ScreenSpot F1@0.5: 80.4% |
| Context Window (Tokens) | 32,000 (32K; previously 2K) |
| License | Open weights; free for personal, research and most commercial use, not for competing paid offerings |
| Multimodality | Yes, vision-language model (image + text; captioning, VQA, detection, pointing, OCR, segmentation) |
| Platform | Cloud API (OpenAI-compatible), local on Mac, Windows, CUDA; Python and Node SDKs |
| Price per 1M Tokens | Cloud real-time: $0.30 input / $1.00 output (Moondream 3.1 9B A2B); batch: $0.15 / $0.50 |
| Release Date | September 18, 2025 (Moondream 3 Preview) |