Home/New releases/NVIDIA Nemotron 3.5 Transcribe
NVIDIA Nemotron 3.5 Transcribe
NVIDIA's open-weight speech model that transcribes 40 languages, labels speakers, translates, and answers questions about a recording.
NVIDIA published Nemotron 3.5 Transcribe on October 28, 2026. One checkpoint handles transcription, speaker attribution (diarization), translation, and audio question answering. It transcribes 40 languages, translates 31 to or from English, and attributes speakers in 13. Recordings can run up to an hour, with up to eight speakers. It has 30B total parameters and about 4B active. The model card says about 4.3B including the audio encoder. It is released in BF16, FP8, and NVFP4. It is built for recorded audio, not live streaming. Audio question answering is trained in English. You can run it with vLLM or SGLang-Omni today. Hosted endpoints are available from Baseten, DeepInfra, and Simplismart. NVIDIA says local deployment with NeMo-Speech.cpp is coming soon. The weights are open under the OpenMDW-1.1 license. NVIDIA's number: first among open-weight models on the Hugging Face Open ASR Leaderboard (English) at 5.33% average WER. NVIDIA also released Granary v2 open speech data manifests, 463,062 hours across 48 languages. Last checked October 29, 2026.
Where it ranks
This is a dated release only. It does not have a rank on a use-case list. Feed label: Voice and audio · Productivity.
Pricing
Check site
Open weights under the OpenMDW-1.1 license. Hosted endpoints are separate. Check the host for a price.
Checked
Added October 29, 2026
Last verified October 29, 2026
Dated releases
NVIDIA released Nemotron 3.5 Transcribe, an open-weight speech model for 40 languages, speakers, translation, and audio questions.