Release dates
— NVIDIA released Nemotron 3.5 Transcribe on 2026-10-28 with OpenMDW-1.1 weights; it has 34B total / approximately 4.3B active parameters and supports transcription across 40 languages and recordings up to one hour. ↗
Nemotron 3.5 Transcribe
An open-weight speech-augmented language model in the Nemotron family. It takes audio and text instructions and returns transcripts, translations, speaker labels, timestamps, or audio-question answers as text.
NVIDIA released BF16, FP8, NVFP4, and GGUF weights on 2026-10-28 under OpenMDW-1.1. The model handles recordings up to one hour in offline inference; the originator has not published a general hosted API model ID.
MODEL NUMBER276
- First public date
- 2026-10-28
- Organization
- NVIDIA
- Family
- Nemotron
01
Specifications
Only published specifications are shown.
- Total parameters
- 34B↗
- Active parameters
- 4.3B↗
- Architecture
- Hybrid↗
- Parameter scope
- The 34B total includes the Transformer audio encoder; the approximately 4.3B active count also includes the joint audio encoder.↗
- Context
- 65,536 tokens↗
- Input modalities
- Audio · Text↗
- Output modalities
- Text↗
- Open weights
- Yes↗
- License
- OpenMDW-1.1↗
- Status
- Current↗
02
Overview
Why it mattered
NVIDIA released Nemotron 3.5 Transcribe as one open-weight model for transcription, translation, and speaker attribution in 40 languages, supporting recordings up to one hour; its model card lists 34B total and approximately 4.3B active parameters. As a specialized speech model, it stays on the timeline. Its unarchived Lightning base is not backfilled and no weight-lineage relation is inferred.
03
Architecture & capabilities
Model capabilities
Audio transcriptionSpeaker diarizationMultilingual
04
Release history
NVIDIA released Nemotron 3.5 Transcribe on 2026-10-28 with OpenMDW-1.1 weights; it has 34B total / approximately 4.3B active parameters and supports transcription across 40 languages and recordings up to one hour.
View source05
Related models
Select a card to view the related model.
No related models recorded yet
06
Sources
01
NVIDIA Nemotron 3.5 Transcribe: Intelligent Transcription with a Speech LLMNVIDIA Nemotron · Published 2026-10-28 · Checked 2026-10-28
Official · Verified02NVIDIA Nemotron 3.5 Transcribe official model card and weightsNVIDIA (official Hugging Face organization) · Published 2026-10-28 · Checked 2026-10-28
Official · VerifiedSuggest a correction
Copy this template, add the field, proposed value, and primary source, and share it with the archive maintainer.