# NVIDIA Open-Sources Nemotron 3.5 Transcribe Speech Model

- Category: Flash (快讯)
- Published: 2026-10-30T12:00:00.000Z
- Source date: 2026-10-30
- HTML: https://www.88z.ai/en/news/flash-20261030-499522b399f7fa9c
- Markdown: https://www.88z.ai/en/news/flash-20261030-499522b399f7fa9c.md
- Other language: https://www.88z.ai/news/flash-20261030-499522b399f7fa9c.md

NVIDIA has released the open-source speech transcription model Nemotron 3.5 Transcribe. Built on a MoE architecture, it supports multilingual transcription, translation, and speaker diarization, ranking among the top open-source models in various benchmarks.

## At a glance

- Uses MoE architecture, supporting transcription in 40 languages and translation between 31 languages and English
- Ranks among the top open-source models in both Open ASR and Artificial Analysis benchmarks

## The update

NVIDIA has officially open-sourced the Nemotron 3.5 Transcribe large speech model. Utilizing a Mixture of Experts (MoE) architecture, it features approximately 34 billion total parameters with 4.3 billion activated. The model can process audio recordings up to one hour long, supporting transcription in 40 languages and translation between 31 languages and English. Additionally, it can distinguish and assign identities to up to eight speakers in real time. In performance benchmarks, the model excels, ranking first among open-weight models in the Open ASR English group with an average word error rate of 5.33%. In the Artificial Analysis non-streaming test, it achieved a word error rate of 2.8%, tying for the top spot among open-source models.

## Sources

- [IT之家](https://www.ithome.com/1/019/545.htm) (2026-10-30)

This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.