About this model
A 30B mixture-of-experts speech recognition model (4B active) for long, noisy, multi-speaker audio.
Official page · AI Search video at 32:04
Every size and format published by NVIDIA, largest first. Want one we don't host yet? Press Request: we add the most requested ones.
| Model | Parameters | Notes | |
|---|---|---|---|
| Nemotron 3.5 Transcribe 30B-A4B | 33.7B | BF16 | |
| Nemotron 3.5 Transcribe FP8 | 33.7B | 8-bit | |
| Nemotron 3.5 Transcribe NVFP4 | 18.7B | 4-bit, Blackwell GPUs |
Nemotron 3.5 Transcribe is an AI voice model by NVIDIA. A 30B mixture-of-experts speech recognition model (4B active) for long, noisy, multi-speaker audio.
Not yet: it can't be hosted yet (license or weights).
License: OpenMDW 1.1. Commercial use of the outputs is allowed by the license.
In the AI Search video at 32:04.
All models · Not affiliated with NVIDIA or with AI Search.