NVIDIA releases open-weight Nemotron 3.5 Transcribe and Granary v2
Nemotron 3.5 Transcribe combines multilingual transcription, translation, speaker attribution and audio question answering in one open-weight model, while Granary v2 provides a public manifest covering 463,062 hours across 48 languages.
AT A GLANCE
Key points
- Transcribes 40 languages and translates 31 to or from English.
- Supports speaker attribution in 13 languages.
- Granary v2 lists 463,062 hours across 48 languages.
What is changing?
NVIDIA released Nemotron 3.5 Transcribe and the Granary v2 speech-data pipeline on October 28. The open-weight model transcribes 40 languages, translates 31 languages to or from English, and supports speaker attribution in 13 languages. NVIDIA also says it combines question answering over audio content in the same model.
Granary v2’s public manifest covers 463,062 hours of audio across 48 languages and 16 source collections. Its reusable preparation pipeline helps teams curate their own recordings for domain adaptation. The model can run on a single GPU and is available for fine-tuning under the OpenMDW 1.1 license; provider benchmarks should still be validated on production data.
Explore the original sources
Source date: 28 October 2026.
Published by AI Radar: 28 Oct 2026, 19:04 (Istanbul, UTC+3).
FOLLOW THE TOPIC
Topics and tags
Share your discovery.
Send this story to a friend.
Reader comments
What do you think about this development? Share your experience or question. Comments are published after editorial approval.
Your name/nickname and comment become public if approved. Read: Privacy notice.
Loading comments…