NVIDIA releases Nemotron 3.5 Super VL and Bolt edge models
NVIDIA launches open-source AI models for agents, ranging from data center servers to local desktop devices, emphasizing speed and efficiency.
Key points
- Nemotron 3.5 Super VL is a 120-billion-parameter model optimized for high-accuracy agentic tasks.
- Two new Bolt models target edge devices, balancing intelligence with memory constraints.
- Organizations are post-training Super VL for specialized finance, legal, and coding workflows.
NVIDIA released Nemotron 3.5 Super VL, introducing an open 120-billion-parameter model designed for specialized always-on agents. The company also announced two new edge-focused models, Nemotron 3.5 Bolt 74B-A9B 24GB and Nemotron 3.5 Bolt E4B 4GB, expanding its open-source portfolio to cover deployments from large enterprise servers down to local hardware.
Super VL delivers leading accuracy on multimodal agentic tasks, including instruction following, tool use, and visual understanding. According to NVIDIA, the model achieves performance comparable to its largest reasoning model while using roughly one-fifth the parameter count. It is described as the fastest open agentic model in its class, operating five times faster than the next best alternative, which helps enterprises complete high-volume tasks spanning days or weeks more quickly.
Context and Background
Unlike closed frontier models, Super VL allows organizations to post-train on their own data, deploy on their own infrastructure, and oversee the development process. The model was built in collaboration with the Nemotron Coalition. It joins existing Nemotron family members, including Lightning, released in August for high-throughput workloads, and Ultra, which handles the hardest reasoning tasks. Bolt brings intelligence to the edge, completing the range of agent deployments.
The two new Bolt models address specific memory and performance needs. Nemotron 3.5 Bolt 74B-A9B 24GB is optimized to balance intelligence and speed per gigabyte for local agentic workloads. It uses native 2-bit quantization to fit in 24GB of memory, allowing it to run alongside other applications on NVIDIA RTX or DGX Spark hardware. Nemotron 3.5 Bolt E4B 4GB is optimized for intelligence per gigabyte, targeting fast conversational agents on GeForce RTX, Jetson, and DGX Spark devices.
Who Is Affected
Dozens of organizations are already post-training Super VL on their own data, tools, and policies. Scale AI post-trained the model for Skydance’s vendor credit finance agent and EY’s M&A due diligence agent. AfterQuery customized it for Hebbia for credit agreement analysis, while AI21 and Alice focused on cybersecurity attack resilience. Mercor post-trained the model on the APEX-SWE coding benchmark for agents that plan, write, and debug code.
Other adopters include Ramp for financial document intelligence, Dream for sovereign AI-powered cybersecurity research, and Lila Sciences for scientific discovery. R1 uses it for medical claims analysis and revenue cycle management. TrainLoop post-trained the model with dermatology data from Nolla, which became the first company approved for AI to write initial patient prescriptions. ModelScope enables developers to fine-tune the model through its ms-swift framework for specialized multimodal applications.
What should you do?
Organizations should evaluate whether their current agent deployments benefit from the reduced parameter count and increased speed of Super VL. Users interested in local execution should test the Bolt 74B-A9B 24GB model on their existing hardware to verify the balance of intelligence and speed. Those deploying conversational agents on edge devices should wait for the October 23 release of Bolt E4B 4GB. Security teams should review post-training data policies, as the open nature of these models allows for custom data integration.
What to do and how to stay safe: NVIDIA
- Audit current agent workloads to identify tasks that could benefit from lower latency and reduced parameter counts.
- Test the Nemotron 3.5 Bolt 74B-A9B 24GB model on available NVIDIA RTX or DGX Spark hardware to assess local performance.
- Review data governance policies before post-training models on proprietary or sensitive organizational data.
- Monitor the October 23 release date for the Nemotron 3.5 Bolt E4B 4GB model if deploying conversational agents on edge devices.
Step-by-step guide: Mobile Device Security: How to Protect Data Beyond Passwords
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
What is the parameter count of Nemotron 3.5 Super VL?
Nemotron 3.5 Super VL is a 120-billion-parameter model.
When is Nemotron 3.5 Bolt E4B 4GB available?
Nemotron 3.5 Bolt E4B 4GB is scheduled to be released on October 23.
Which organizations are post-training Nemotron 3.5 Super VL?
Organizations including Scale AI, AfterQuery, AI21, Alice, Mercor, Ramp, Dream, Lila Sciences, R1, and TrainLoop are post-training the model.
