Small models, big vision
Moondream is the flagship of M87 Labs, founded by Jay Allen (CEO) and Vik Korrapati (CTO), the developer who created the original moondream model. The company emerged from stealth in October 2024 with $4.5M in pre-seed funding led by Felicis, joined by Microsoft's M12 GitHub Fund and Ascend, on a contrarian thesis: for real-world computer vision, smaller is better.
The stack
Moondream ships open-weight vision-language models at 0.5B, 1B and 2B parameters that handle image captioning, visual question answering, object detection and pointing — competitive with far larger models on benchmarks like VQAv2 while running on phones, Raspberry Pis and edge hardware. Around the models, M87 Labs offers Lens, a fine-tuning platform that works from as few as 20 labeled images; Photon, a speed-optimized inference engine; and Moondream Cloud, a hosted API priced at $0.06 per 1,000 images with $5 in free monthly credits.
Traction
With thousands of GitHub stars and adoption across robotics, retail analytics and media workflows, Moondream has become the default open VLM for developers who need production vision AI without GPU clusters.