Prime Intellect Acquires Moondream to Integrate Photon Inference Engine

Prime Intellect has acquired Moondream, bringing founders Vik Korrapati and Jay Allen into the company to lead inference scaling and multimodal model development. The acquisition integrates Moondream’s Photon inference engine and custom GPU kernels into Prime’s stack, aiming to combine efficient serving with Prime’s training platform.

Performance and Efficiency

Moondream’s technology focuses on small, specialized models that outperform larger generalists on specific tasks. Moondream 3.1 leads in object detection and chart analysis, achieving 34 requests per second, which is approximately 24x faster than the next most accurate open model. Compared to Meta’s Muse Wonder, Moondream 3.1 reaches 80% of its pointing accuracy at roughly 1/1000th of the cost.

The Photon engine compiles models into single GPU programs, enabling answers in tens of milliseconds. On a single NVIDIA H100, Photon runs more than 60 Moondream inferences per second, supporting live video applications.

Availability and Integration

Moondream 3.1 is now available through Prime Inference, powered by Photon. Additionally, Clef and Clef Flash models are live on Prime Inference and OpenRouter, with Clef priced at parity with Jev and Clef Flash at half the price. Existing Moondream Cloud customers will receive platform credits usable for Moondream models and broader training offerings. Moondream will remain the brand for small, efficient multimodal models within Prime’s ecosystem.