GIGABYTE says its AI TOP ATOM hardware platform fully supports the newly released NVIDIA Nemotron 3.5 Lightning model and the NVIDIA NeMo Switchyard open source routing library, aiming to give customers on-premises agentic AI with high performance and full data sovereignty.

The model is built for the high-volume, repetitive execution steps in always-on agents, such as personal assistants, software development, cybersecurity operations, financial and healthcare administration, and telecom and retail workflows. NeMo Switchyard selects the right model for each agent step based on workflow requirements.

Key Features at a Glance

• Throughput: Up to 4x higher throughput than same-class models – shorter response times across complex multi-turn conversations and automated workflows.
• Architecture: Hybrid Mixture-of-Experts (MoE) with 30B parameters and 3B active parameters, distilled from NVIDIA's frontier Nemotron 3 Ultra model.
• Out-of-the-Box Accuracy: Trained for popular agent harnesses, with industry-leading accuracy for coding, tool calling, and instruction following.
• Context Length: Up to 1M context – handles long-context retrieval and multi-turn workflows.
• Control and Privacy: Open model that can be deployed locally, at the edge, or in the cloud, with domain-specific post-training; sensitive data stays within the local network during training and inference.
• Hardware Range: Runs on NVIDIA Jetson, RTX and RTX PRO, DGX Spark and DGX Station, and datacenter servers including HGX H100, HGX H200, GB200 NVL72, and GB300 NVL72 – so agents can run anywhere.
• Desktop Option: AI TOP ATOM pairs with the NVIDIA GB10 Grace Blackwell Superchip so developers can deploy, post-train, and run the model at their desks.

Availability

The announcement gives no release date or pricing; GIGABYTE says it offers immediate support for the Nemotron 3.5 Lightning model and NeMo Switchyard on both datacenter servers and the AI TOP ATOM desktop platform.

GIGABYTE AI TOP ATOM and Nemotron 3.5 Lightning Technical Details