NVIDIA-Nemotron-3.5-Transcribe-30B-A4B (nvidia) specs, prices, where it runs and its benchmark standing

Facts

Provider

Nvidia

Release date

Oct 7, 2026

Hugging Face

Parameters

33.7B total, 4B active per token

Hugging Face

Modalities

Audio to Text

Hugging Face

License

OpenMDW 1.1

Hugging Face

Hosting

self-host: weights on Hugging Face

Hardware

Estimate: about 67 GB of weights as published (BF16), about 34 GB at 8-bit, about 17 GB at 4-bit, plus memory for the context

Hugging Face config

Benchmarks

1 board captured