Nvidia Nemotron 3.5 Transcribe Brings 40 Language Voice AI to One Chip
Nvidia’s free Nemotron 3.5 Transcribe supports 40 languages, hour-long recordings and eight speakers on one chip, but cannot process live audio streams.
Summary
Nvidia released Nemotron 3.5 Transcribe on Wednesday, October 28, 2026, alongside version 2 of its open source Granary speech dataset. The free, open-weight, 30 billion parameter model transcribes up to 40 languages, translates 31 languages to and from English, processes recordings up to one hour, and distinguishes up to eight speakers. Nvidia claims it outperforms rivals on benchmarks including Hugging Face’s Open ASR Leaderboard, but it handles recorded audio rather than livestreams.
Developers can combine Nemotron 3.5, Granary and NeMo Speech to customize business meeting and medical transcription apps for industry jargon across languages. Nvidia says the model runs on one chip, spanning the Jetson AGX Thor robotics controller, DGX Spark hardware powering new laptops, and Blackwell datacenter chips. It joins Nvidia’s smaller 1.1 billion parameter Parakeet voice model. Competition is accelerating: Mistral shipped Voxtral 3 earlier in October, ElevenLabs launched Realtime for instant speech in voice agents, and Salesforce released Launa-Omni.
Positives
- Nemotron 3.5 Transcribe supports up to 40 languages and translates 31 languages to and from English.
- Recordings up to one hour can be separated across as many as eight speakers.
- A 30 billion parameter model can run on one Jetson AGX Thor, DGX Spark or Blackwell chip, Nvidia says.
- Granary version 2 and NeMo Speech support customization for multilingual industry jargon.
- Free, open-weight access gives business meeting and medical transcription developers another adaptable voice AI option.
Risks & concerns
- Recorded audio is supported, but Nemotron 3.5 Transcribe cannot process livestreams.
- One-hour recording support imposes a stated duration ceiling.
- Speaker separation is capped at eight participants.
- Mistral, ElevenLabs and Salesforce released competing voice products during the same busy period.
- Nvidia’s claimed benchmark lead includes Hugging Face’s Open ASR Leaderboard but remains presented as a company claim.