I noted NVIDIA's announcement of the Nemotron 3.5 Super VL model, highlighted as a new open model that integrates reasoning, tool use, and multimodal understanding across text, image, document, and video, aimed at long‑running AI agents.
The model has 120 B total parameters.
It uses 12 B active parameters per inference via a sparsely activated architecture.
Gone are the days of training a custom YOLO for every use case. (Nemotron 3.5 Super VL + Hermes Agent) https://t.co/Or9mCQ8Qzq