NVIDIA Releases Nemotron
NVIDIA keeps expanding its open-weights portfolio, and the strategy looks deliberate: give away capable models, sell the compute to run them. A 4.71% mean WER on the Open ASR Leaderboard puts this at the top tier of English ASR, and the MoE design with only ~4B active parameters makes it cheap enough for real-time deployment.… more
Speaker labels and word timestamps are the underrated part — that is what turns a transcription model into a production component for meeting bots and call-center analytics. The catch: the headline benchmark is English-only, and quality across the other 39 languages is the real test. Still, Whisper's de facto monopoly on open speech models is clearly under attack, and enterprises get another credible self-hostable option.
NVIDIA Releases Nemotron 3.5 Transcribe: An Open 30B MoE Model for 40-Language Transcription with ~4B Active Parameters
NVIDIA released Nemotron 3.5 Transcribe, an open-weights speech LLM with 4.3B active parameters. It transcribes 40 languages, adds speaker labels and word timestamps, and posts a 4.71% mean WER on the English Open ASR Leaderboard…
Source:
Marktechpost
Discussion (0)