This paper speeds up how AI models read very long texts by carefully choosing which words (tokens) to focus on at each step.
Nemotron 3 is a new family of open AI models (Nano, Super, Ultra) built to think better while running faster and cheaper.