Prime Intellect acquires Moondream for faster inference

This is a strategic acqui-hire aimed at making open multimodal models cheaper and faster to serve, especially for real-time perception and computer-use workloads.

• –Prime gains a compact vision-model portfolio plus specialized inference expertise.
• –Photon’s compiled megakernel approach targets lower latency and less CPU-GPU overhead than conventional serving stacks.
• –The acquisition strengthens Prime’s position in vision, robotics, live video, and computer-use infrastructure.
• –The 60-inferences-per-second figure is vendor-reported and based on internal H100 measurements, so developers should validate performance on their own workloads.
• –Moondream remains a brand, while Photon and its kernel technology are being integrated into Prime’s broader training and inference platform.