NVIDIA releases Nemotron 3.5 Super VL and Bolt, a 24GB open model for local agents
Announced · Published · By Contractor AI Shop
NVIDIA released Nemotron 3.5 Super VL, an open 120B multimodal model with 12B active parameters for always-on agents, and Bolt 74B-A9B, packed into 24GB with 2-bit quantization to run local agents. A smaller Bolt E4B 4GB model is due Oct. 23.
Key facts
- NVIDIA says Bolt 74B-A9B uses native 2-bit quantization to fit in 24GB and run alongside other applications. NVIDIA
- NVIDIA says Bolt is on Hugging Face with support for llama.cpp, Ollama, LM Studio, Unsloth and vLLM. NVIDIA
- The Decoder reports Super VL is an open model built to run enterprise AI agents continuously. The Decoder
- SiliconANGLE reports NVIDIA expanded the Nemotron family with efficiency-focused models for agentic systems. SiliconANGLE
Our take: what it means for contractors
Bolt is the one to watch: a 24GB footprint is within reach of a well-equipped Mac or a 24GB GPU through Ollama or LM Studio. Test it on your own specs and RFIs before trusting it; Super VL is a data-center model.
What we don't know yet
- Real speed on Apple Silicon (NVIDIA's figures are on RTX Spark)
- How much memory vision and the speculative drafter add beyond 24GB
- Whether Bolt E4B ships Oct. 23 as stated
Sources
- NVIDIA · Open From Data Center to Desktop: NVIDIA Expands Nemotron Family With Highest Efficiency Super VL and Bolt Models · 2026-10-15 · Primary
- The Decoder · Nvidia's Nemotron 3.5 Super VL is an open model built to run enterprise AI agents nonstop · 2026-10-15 · Secondary
- SiliconANGLE · Nvidia expands Nemotron LLM family with hyper-efficient models for agentic systems · 2026-10-15 · Secondary