Nvidia Releases Nemotron 3.5 for Up to 40 Languages — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Nemotron 3.5 Transcribe was released Wednesday as a free, open-weight model that transcribes audio in up to 40 languages.
- Nemotron 3.5 Transcribe can translate 31 languages to and from English, handle recordings up to one hour long and distinguish up to eight speakers.
- Nvidia claims Nemotron 3.5 Transcribe beat rivals on benchmarks including Hugging Face's Open ASR Leaderboard; the tools target business-meeting and medical-transcription apps, two established workplace footholds for AI.
- Granary version 2 was released as an open-source speech dataset; developers can use it with Nemotron 3.5 Transcribe and NeMo Speech to customize apps that recognize industry jargon in each language.
- Nemotron 3.5 Transcribe accepts recorded audio but not livestreams, a limitation Nvidia identifies in the release.
- Nemotron 3.5 Transcribe is a 30-billion-parameter model that Nvidia says can run on one chip, including Jetson AGX Thor, DGX Spark and Blackwell datacenter chips; it joins the 1.1-billion-parameter Parakeet in Nvidia's open-weight voice lineup.
- Nvidia launched Nemotron into a busy AI-voice release cycle: Mistral shipped Voxtral 3 earlier this month, ElevenLabs launched Realtime for instant speech in voice agents, and Salesforce released Launa-Omni.
Why it matters: For developers building meeting and medical-transcription apps, one Nvidia stack now combines free, open-weight 40-language transcription, one-hour recording support, eight-speaker separation and single-chip execution. That concentration of capabilities has a hard boundary: the model accepts recorded audio only, leaving livestream workflows unsupported.
Ask SkimNews
