Induction Labs Photon-1 Simulates Desktops, Plays Checkers, and Models Billiard Physics From One Pretraining Run

Induction Labs unveiled Photon-1, a large MoE model pretrained on raw computer-use video without action labels, then fine-tuned for desktop control.

The company says Photon-1 uses imagination-style pretraining and next-latent-token prediction to learn useful policies from video alone.

On internal benchmarks, it reportedly outperformed Gemini 3.1 Flash-Lite while training and serving at much lower cost.

If validated, the approach could reduce reliance on expensive action-labeled data and make computer-use agents cheaper to build and deploy.