NVIDIA Expands Nemotron Family With Open Super VL, Bolt Models

NVIDIA released the open-weight model Nemotron 3.5 Super VL on October 15 and introduced two Bolt models designed for local devices. Super VL features 120 billion total parameters, combining text and visual understanding capabilities to handle documents, charts, tool calls, and multi-step business workflows. NVIDIA claims it offers a speed advantage in agentic tasks among comparable open models, though this performance conclusion should be verified against specific testing conditions. The Nemotron 3.5 Bolt 74B-A9B 24GB employs quantization technology to run with approximately 24GB of memory on devices equipped with corresponding hardware, targeting local agentic tasks such as web research, programming, and debugging. The smaller Bolt E4B 4GB targets voice assistants, interactive characters, and embedded applications. Super VL and Bolt 74B-A9B are currently available, while Bolt E4B is scheduled for release on October 23. NVIDIA stated that Super VL provides model weights, datasets, and training schemes, allowing enterprises to perform continued training with their own data and select deployment environments based on security and data residency requirements. Organizations such as Scale AI and Ramp are using it for specific business scenarios. The combination of the three models reflects NVIDIA’s aim to cover deployment needs across data centers, desktop devices, and edge hardware within a single model family.