Nemotron 3.5 brings open multimodal agents from servers to RTX PCs
NVIDIA’s open Nemotron 3.5 lineup gives organizations a 120-billion-parameter multimodal agent model for enterprise servers, plus smaller models for local NVIDIA hardware. Super VL handles visual understanding, tool use and multistep workflows, and NVIDIA says it runs 5x faster than the next-best open agentic model in its class. Bolt 74B-A9B uses 2-bit quantization to fit in 24GB, so local agents can research, code and debug without sending workloads to the cloud.