Nvidia releases open Nemotron 3.5 Transcribe speech model

Nemotron 3.5 Transcribe is described as an open-weight 30B model supporting transcription, diarization, translation, and audio Q&A across 40 languages, alongside the Granary v2 dataset. Nvidia claims benchmark leadership; verify weights, licensing, evaluation conditions, and the near-field/experimental limitation on speaker transcription.