NVIDIA drops Nemotron 3.5 ASR
NVIDIA has released Nemotron 3.5 ASR, a 600M-parameter cache-aware streaming Automatic Speech Recognition (ASR) model. Featuring multilingual support for 40 language-locales from a single checkpoint, native punctuation, and low latency, it is fully open-source and fine-tunable, enabling local deployment across various hardware.
While proprietary ASR APIs have dominated, NVIDIA is enabling the next wave of local, real-time voice applications by open-sourcing a highly optimized, low-latency model.
* FastConformer-RNNT architecture supports cache-aware streaming for sub-100ms latency.
* Single checkpoint handles 40 language-locales natively with punctuation and capitalization.
* Highly accessible 600M-parameter size allows hosting on standard laptop CPUs up to enterprise GPUs.
DISCOVERED
1h ago
2026-06-08
PUBLISHED
2h ago
2026-06-08
RELEVANCE
AUTHOR
anarchyco