Nemotron 3.5 Bolt (74B A9B 24GB)

NVIDIA Nemotron 3.5 Bolt is an open multimodal 74B total / 9B active Mixture-of-Experts vision-language model for local / edge agentic workloads. Compressed from Nemotron 3.5 Super VL into a native IQ2 (2-bit) 24GB footprint for RTX / DGX Spark and similar systems, with a hybrid Mamba-2 + LatentMoE + Attention backbone and C-RADIO v4-H vision encoder. Accepts text, image, and video inputs and produces text; up to 256K context; tool use and configurable reasoning. A full-precision BF16 sibling is published at nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16. Released under OpenMDW-1.1.