Breaking

NVIDIA Introduces Nemotron 3.5 Transcribe for Multilingual Audio

October 28, 2026 at 11:18 EDT

Foundation Models
Open Source

NVIDIA has introduced Nemotron 3.5 Transcribe, an open-weight 30B mixture-of-experts model with about 4B active parameters, bringing transcription, speaker attribution, translation, contextual prompting and audio understanding into one checkpoint. The model is described as supporting 40 languages.

October 28, 2026 · NVIDIA

One checkpoint for a wider range of speech tasks

Nemotron 3.5 Transcribe combines multilingual transcription with speaker attribution, translation, contextual prompting and audio understanding.

supported languages

shared checkpoint

Model capacity and active compute

Each stacked block represents 1 billion parameters.

A mixture-of-experts model activates only part of its parameter set for a given pass.

Unified task set

Transcription, translation, speaker attribution and audio understanding in one checkpoint.

Serving gap

The current SGLang path handles uploaded files, but lacks live PCM input, translation and speaker-attributed output.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.