1
Moondream 3 Preview: Frontier-level reasoning at a blazing speed
Think deeper, detect smarter, run just as fast.
A fast & powerful vision model that rocks.
Visit moondream.ai → On Hacker News Sep 2025 – Jun 2026
Think deeper, detect smarter, run just as fast.
Photon, Moondream's inference engine, achieves near-realtime VLM inference (~33ms on NVIDIA B200). This is a peek into how it delivers up to 35% higher decode throughput by optimizing how the GPU works.