Compliance
Nvidia Launches Nemotron 3.5 ASR and Bolt E4B Models for Multilingual Transcription and Local AI
Nvidia has introduced two significant models in its Nemotron 3.5 lineup: the Nemotron 3.5 ASR for audio transcription and the Nemotron 3.5 Bolt E4B for local voice assistants.
Nemotron 3.5 ASR Details
The Nemotron 3.5 ASR is a free, open-weight speech model capable of transcribing audio in up to 40 languages. It also functions as a translator for 31 languages to and from English. Key features include:
- Handles recordings up to one hour long.
- Can distinguish between up to eight speakers.
- Supports real-time streaming with low latency of 0.07 seconds.
- Built on a cache-aware FastConformer-RNNT architecture.
This model is available on Hugging Face under the OpenMDW-1.1 license.
Bolt E4B Model Overview
Launched on October 23, 2026, the Nemotron 3.5 Bolt E4B is designed for running voice assistants and AI agents directly on local hardware. Key specifications include:
- 4GB model optimized for low-memory hardware like GeForce RTX GPUs and Jetson edge devices.
- Post-trained configuration of Google’s Gemma 4 E4B, featuring 8 billion total parameters.
- Approximately 40% higher benchmark performance than its predecessor.
- Supports 35 languages and efficient multi-turn conversations.
Both models are part of Nvidia's strategy to enhance AI capabilities across various applications, with open access to weights and datasets for developers.
FAQ
What is the primary function of the Nemotron 3.5 ASR model?
The Nemotron 3.5 ASR model is designed for audio transcription, capable of transcribing audio in up to 40 languages and translating between 31 languages and English.
What are the key features of the Nemotron 3.5 ASR?
Key features include handling recordings up to one hour long, distinguishing between up to eight speakers, supporting real-time streaming with low latency of 0.07 seconds, and being built on a cache-aware FastConformer-RNNT architecture.
When was the Nemotron 3.5 Bolt E4B model launched?
The Nemotron 3.5 Bolt E4B model was launched on October 23, 2026.
What hardware is the Bolt E4B model optimized for?
The Bolt E4B model is optimized for low-memory hardware, such as GeForce RTX GPUs and Jetson edge devices.
How does the performance of the Bolt E4B compare to its predecessor?
The Bolt E4B model offers approximately 40% higher benchmark performance than its predecessor.