NVIDIA today announced the release of the Nemotron 3.5 Super VL and two Bolt models, expanding its Nemotron family of AI models NVIDIA Newsroom.
The Nemotron 3.5 Super VL is a 120-billion-parameter model designed for specialized always-on agents, delivering what NVIDIA claims is the fastest task completion and highest accuracy in its class. The model is optimized for multimodal agentic tasks, including instruction following, tool use, multistep business workflows, and visual understanding. NVIDIA states that Super VL achieves leading accuracy comparable to its largest reasoning model but with roughly one-fifth the parameter count. The model is available now on platforms like Hugging Face, Open Router, and ModelScope community, as well as NVIDIA NIM.
Joining Super VL are two new edge models: Nemotron 3.5 Bolt 74B-A9B 24GB and Nemotron 3.5 Bolt E4B 4GB. The Bolt 74B-A9B 24GB is optimized for local agentic workloads, balancing intelligence and speed per GB. It is designed to run on NVIDIA RTX/DGX Spark and is available now. The Bolt E4B 4GB, optimized for intelligence per GB, is set to be released on October 23 and is intended for fast conversational agents on GeForce RTX, Jetson, and RTX/DGX Spark.
NVIDIA claims that these models cover the full range of agent deployments, from large enterprise servers to edge devices and local hardware. The company also highlights that organizations can post-train Super VL on their own data, deploy it on their own infrastructure, and oversee what they build. This openness and customizability are reinforced by the model's open weights, open datasets, and training recipes.
The Nemotron 3.5 Super VL has been adopted by various organizations across industries for specialized tasks. For example, Scale AI has post-trained the model for Skydance’s vendor credit finance agent and EY’s M&A due diligence agent. Other organizations, such as AfterQuery, AI21, and Alice, have customized the model for tasks like credit agreement analysis and cybersecurity attack resilience. The model is also being integrated into software platforms like Atlassian, CrowdStrike, and Salesforce for tasks ranging from agentic search to enterprise security and engineering design.
NVIDIA states that the Nemotron 3.5 Super VL was trained on about 26 trillion tokens across pretraining and continued pretraining, with roughly 3.8 million samples and reinforcement learning across dozens of environments. The model is available on NVIDIA’s full hardware stack, including DGX Spark, DGX Station, RTX PRO, and RTX systems, as well as NVIDIA Jetson at the edge and across data center and cloud environments.
The announcement does not include independent test results or benchmarks to verify NVIDIA's claims about the models' performance and accuracy.