Nvidia launches a promptable transcription model

Nvidia released Nemotron 3.5 Transcribe, a 30B mixture-of-experts model that allows users to set context and instructions before transcription. The model achieves a 5.33% average word error rate on English and tops speaker-attribution benchmarks. It supports translation, verbatim or cleaned transcripts, and custom formatting. The tradeoff is significantly larger model size compared to traditional sub-billion parameter transcription systems.

Summary and key points written by AI from the headline and a short excerpt, so they can be incomplete or wrong. The New Stack did the reporting; read the original for the full story.