Nova is the world's deepest-trained ASR model to date, achieving unrivaled performance across diverse automatic speech recognition tasks. Nova delivers a 22% reduction in WER, 23-78x faster inference, and 3-7x lower cost than its closest competitor.