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, supporting 40 languages and achieving 4.71% mean WER on the English Open ASR Leaderboard.
Key Takeaways
- Key Highlight：NVIDIA released Nemotron 3.5 Transcribe, an open-weights speech LLM with 4.3B active parameters, supporting 40 languages and achieving 4.71% mean WER on the English Open ASR Leaderboard.
- Key Topics：NVIDIA, Releases, Nemotron, An
NVIDIA has released Nemotron 3.5 Transcribe, a new open-weights speech model built on a mixture-of-experts architecture. It has roughly 30 billion total parameters but only about 4.3 billion active per token, making it efficient for inference. The model handles transcription across 40 languages and also outputs speaker labels and word-level timestamps.
This release adds to the open speech-recognition space by combining strong accuracy with a relatively small active parameter count. The reported 4.71% mean word error rate on the English Open ASR Leaderboard suggests it is competitive with larger or closed models. By releasing weights openly, NVIDIA is giving developers a flexible base for transcription, diarization, and timestamping tasks.
The model could lower barriers for building multilingual voice applications, especially those needing speaker attribution and precise timing. It also fits NVIDIA's broader strategy of releasing open models that run well on its hardware. However, the practical value will depend on how it performs on real-world audio, not just leaderboard benchmarks.
This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.