NVIDIA's distilled Cosmos 3 Super model generates video in 4 steps instead of 50, topping open-weights leaderboards with up to 25x faster inference.
Continue to AI University →