NVIDIA releases open model Nemotron 3.5 Bolt E4B 4GB for fast voice and embedded assistants
NVIDIA has released the open language model Nemotron 3.5 Bolt E4B 4GB, derived from the Gemma-4-E4B-it model, intended for local voice and embedded assistants. According to the company, it achieves the highest average accuracy on 21 benchmarks with the smallest number of output tokens.
NVIDIA has released the Nemotron 3.5 Bolt E4B 4GB model, an open small language model created by further fine-tuning the Gemma-4-E4B-it model. The model focuses on speed and memory efficiency for conversational agents and, according to the company, extends capabilities for multi-turn tool use, instruction following, and input context processing.
The model is intended for local voice and device assistants, interactive virtual characters, and embedded/edge interfaces running on GeForce RTX, Jetson, and RTX/DGX Spark hardware. According to NVIDIA, it achieved the highest average accuracy on 21 benchmarks while generating the smallest number of output tokens compared with the Gemma-4-E4B-it and Qwen3.5-4B models, both in reasoning mode and without it.
The model is released together with the DFlash drafter accelerator, which, according to the company, speeds up single-user decoding 2.74 times compared to inference without speculative decoding at concurrency one, and achieves 1.21 times higher generation throughput than the Gemma-4-E4B-it model with MTP at batch size one.
Why it matters
For developers of local voice assistants and embedded devices (IoT, robotics), it offers a smaller open model that can be run directly on the device without a cloud connection, with lower latency. Companies developing products on NVIDIA hardware (Jetson, RTX) thus gain a reference model optimized directly for these platforms.
Release card
Nemotron 3.5 Bolt E4B 4GB
NVIDIA
- Rychlost dekódování na jednoho uživatele s akcelerátorem DFlash drafter 2,74x At concurrency one, decoding with the DFlash drafter accelerator is 2.74 times faster than inference without speculative decoding.
- Propustnost generování s akcelerátorem DFlash drafter 1,21x At batch size one, the model with the DFlash drafter accelerator achieves 1.21 times higher generation throughput than the Gemma-4-E4B-it model with MTP.
- Local voice and device assistants
- Interactive virtual characters
- Embedded and edge interfaces on GeForce RTX, Jetson, and RTX/DGX Spark hardware
The model is compared with the Gemma-4-E4B-it and Qwen3.5-4B models, against which, according to the source, it achieves the highest average accuracy on 21 benchmarks while generating the smallest number of output tokens, both in reasoning mode and without it.
The card summarizes information from the article and any dated corrections, with a link to the original source. It is not our assessment of the model. It does not yet have a dedicated editorial profile. Model selection and other announcements →
Two audiences, two different impacts
What this means
For individuals
Developers and enthusiasts building local assistants or embedded AI projects have access to a smaller open model that, according to the company, can be deployed directly on devices such as Jetson or RTX without needing a cloud API.
For a business
Companies developing voice assistants, robotics, or IoT products on NVIDIA hardware can, according to available information, deploy a smaller open model locally, which reduces dependence on cloud APIs.
DevelopmentCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.