Nvidia launches Nemotron 3.5 Super VL, a 120B-parameter open model for enterprise agents, and says it outperforms frontier models on specialized benchmarks

Enterprise agents can now leverage multimodal understanding and faster execution without sending sensitive data to the cloud.

NVIDIA has introduced three new open-weight AI models: the 120-billion-parameter Nemotron 3.5 Super VL for enterprise agents, and two smaller edge models, Nemotron 3.5 Bolt 74B-A9B (24GB) and Nemotron 3.5 Bolt E4B (4GB). The Super VL model is designed for specialized, always-on agents, claiming leading accuracy on multimodal tasks like instruction following, tool use, and visual understanding, performing comparably to larger models at a fraction of the parameter count. It boasts robust visual capabilities for extracting data from documents and can be fine-tuned by enterprises on proprietary data for enhanced privacy. The Bolt models are optimized for local agentic workloads and conversational agents, with the 74B-A9B model utilizing 2-bit quantization to fit into 24GB of memory for local task management, and the E4B model designed for fast physical and conversational agents on consumer hardware.

NVIDIA's release of the Nemotron 3.5 Super VL, a 120B-parameter open model, signifies a strategic move to empower enterprise AI agents with advanced multimodal capabilities and self-sovereignty. By offering an open-weight model that can be fine-tuned on proprietary data, NVIDIA addresses critical concerns around data privacy and control for businesses. The model's claimed performance parity with larger, closed models, at a significantly reduced parameter count and increased speed, suggests a potential inflection point in the efficiency and accessibility of sophisticated AI agents for business workflows.

The introduction of the Nemotron 3.5 Bolt models further emphasizes NVIDIA's commitment to pushing AI to the edge and enabling localized agentic operations. The 2-bit quantization in the 74B-A9B model, allowing it to run on 24GB of memory, is a key enabler for cost-effective deployment and on-device processing, catering to a wider range of enterprise hardware. This focus on hybrid and edge deployments indicates a market trend towards decentralized AI, where specialized models are optimized for specific tasks and environments, reducing latency and dependence on cloud infrastructure.