Summary
On October 15, 2026, NVIDIA released Nemotron 3.5 Super VL, a 120B-parameter open multimodal mixture-of-experts model (12B active per token) built to run enterprise agents continuously, alongside Bolt models for local hardware. Super VL and Bolt 74B-A9B (which fits in 24GB via 2-bit quantization) are available now through Hugging Face, OpenRouter, ModelScope, and NVIDIA NIM, with the smaller Bolt E4B for voice assistants arriving October 23.
What changed
NVIDIA added three models to the Nemotron family: Super VL (120B MoE, 12B active, multimodal, open weights under OpenMDW-1.1) for always-on enterprise agents, and Bolt 74B-A9B (24GB local footprint) plus the smaller Bolt E4B (arriving Oct 23) for on-device and voice agents, distributed via Hugging Face, OpenRouter, ModelScope, and NIM.
Why it matters
Open, efficient agent models that run on local hardware or in-house servers let enterprises operate agents without constant cloud inference, cutting cost and data exposure for sensitive workloads and strengthening the open-weight alternative to closed frontier APIs.
Evidence excerpt
Super VL and Bolt 74B-A9B are available now, with Bolt E4B coming Oct. 23.