SAP targets efficiency gains above €2 billion with cheaper AI
The software company is changing models in some products while expecting its overall AI spending to rise as adoption expands.
Nvidia launches open-weight transcription model covering 40 languages
Nemotron 3.5 Transcribe is free and works with recorded audio rather than live streams. Nvidia also released an updated speech dataset.
Nvidia released Nemotron 3.5 Transcribe on October 28, 2026, offering developers a free, open-weight audio transcription model with support for up to 40 languages. Alongside it, the company introduced Granary version 2, an update to its open-source speech dataset.
The tools are aimed at developers creating applications for medical transcription and business meetings. Nvidia said they can combine the model and dataset with tools such as NeMo Speech to adapt applications to industry-specific vocabulary across languages. Transcribe processes recordings, but does not support live audio streams.
According to Nvidia, the model supports translation between English and 31 languages, processes hour-long recordings and distinguishes as many as eight speakers. The company also claimed stronger benchmark results than competing models, including on Hugging Face’s Open ASR Leaderboard.
Nvidia said the 30 billion-parameter model can operate on a single chip. It expands the company’s open-weight voice lineup, which includes the 1.1 billion-parameter Parakeet model. Other recent voice AI launches include Mistral’s Voxtral 3, ElevenLabs’ Realtime for voice agents and Salesforce’s Launa-Omni.