Back to the stories

NVIDIA releases Nemotron 3.5 Super VL open model for specialized always-on agents

Score 9.2,

Artificial Intelligence

Building on our coverage from three days ago, NVIDIA now positions Nemotron 3.5 Super VL as an open model for specialized always-on agents and has bundled it with production-ready inference tooling.

NVIDIA first announced Nemotron 3.5 Super VL on October 15 and materially updated that announcement on October 19. The company describes the release as an open model intended for agents that run continuously to complete tasks quickly and accurately. NVIDIA also ships the model with NIM: optimized inference engines, validated configurations, and standard APIs for cloud and data-center deployment.

Why this matters: until now, open models were mostly research artifacts that teams had to adapt and tune themselves. This package changes that. Teams get not just a model, but a tested deployment stack that shortens the path from prototype to a reliable, always-on service.

How it works: Nemotron 3.5 Super VL is a 120 billion-parameter multimodal reasoning model. Think of it as a powerful engine sold with a ready-to-install transmission and wiring. The model is the engine; NIM provides the tuned runtime, the installation instructions and the connectors that make the engine run continuously in a data center.

What changes now: developers and companies building task-specific agents for monitoring, scheduling or customer support can move faster. Startups that relied on hosted APIs can experiment on their own infrastructure instead of depending on a third party. That said, running this stack still requires data-center class GPUs; NIM lowers integration work but does not make the model desktop-ready.

What to watch next: independent benchmarks and the general availability rollout. The key question is whether teams will actually use this open, production-ready offering to replace hosted agent services, or mostly to prototype before returning to managed APIs.