According to Nvidia's blog, the model takes plain-language prompts and performs transcription, speaker identification and translation together, and can also answer questions about what was said in a recording. It transcribes 40 languages, including seven Indic languages. TechCrunch reports that it translates 31 languages to and from English, handles recordings up to one hour long and distinguishes up to eight speakers. Prompts can include product terms and formatting conventions to handle specialized vocabulary in fields such as healthcare, financial services and customer service.
TechCrunch reports the model has 30 billion parameters; Nvidia says about 4.3 billion of them are active at a time. The company claims it ranks first on Hugging Face's Open ASR leaderboard and second among open-weight models on Artificial Analysis. It ships in BF16, FP8 and NVFP4 formats and runs on a single chip, from RTX Spark and DGX Spark machines and Jetson AGX Thor devices up to Blackwell datacenter chips. Developers can adapt it to industry jargon using Granary 2 and the NeMo Speech tools.
The announcement came with an update on RTX Spark laptops, which are now available from ASUS, Dell, HP, Lenovo, Microsoft and MSI starting at $2,599. According to TechCrunch, the model works on recorded audio only, not live streams. Full license terms, the complete language list and detailed accuracy scores were not given in the announcement.
