AnalysisAI ModelsSeptember 4, 2026

WorldReward: Reward model for camera-conditioned world models

WorldReward is a vision-language reward model that evaluates camera-conditioned world models by aligning video chunks with actions and aggregating preferences for execution consistency and visual quality. It introduces WorldReward-Bench and a reasoning-augmented preference dataset for RL post-training.

1 source

More stories today

Open the live feed
WorldReward: Reward model for camera-conditioned world models