Nvidia Expands Nemotron With Open Models for AI Agents
The new Nemotron 3.5 models target enterprise, multimodal and local agent workloads across data-centre and smaller devices.
Nvidia has expanded its Nemotron family with new open models designed for specialised and local AI agents.
What happened
The release includes Nemotron 3.5 Super VL, a 120-billion-parameter multimodal model for complex agent tasks, and two smaller Bolt models designed to balance capability, speed and memory use on local hardware. One Bolt model fits within 24GB of memory, while a smaller 4GB version targets lightweight conversational agents.
Why it matters
Running agents locally can reduce latency, cloud costs and data exposure. Smaller open models also give companies more control over how systems are deployed and adapted for specific workflows.
The bigger picture
Nvidia is extending beyond chips into the models and software that determine how its hardware is used. Its position in AI infrastructure increasingly covers the full stack, from processors to deployable agent systems.