News/Open Source
Try nowOpen Source·IncrementalOfficialbreakingUpdated Oct 15·Updated 1×·First seen Oct 15
NVIDIA Releases Nemotron 3.5 Super Vl and Bolt for Edge Agents
LaunchOpen source
Try it
Something you can actually use or run today.
On October 15, 2026, NVIDIA released the Nemotron 3.5 Super VL (120B parameter VLM) and two edge-optimized Bolt models (74B and 4B).
It provides high-intelligence vision models that can fit on single-consumer GPUs (24GB VRAM) for 'always-on' edge applications.
AILookup take
NVIDIA is building the models that sell its hardware. Nemotron Bolt is the real win here—high-accuracy vision on 24GB VRAM is the sweet spot for small-scale automation.
Who cares
IoT developersedge computing engineerscomputer vision researchers
Watch next
Integration of Bolt models into standard open-source vision pipelines like ComfyUI or Home Assistant.
Details
- Bolt is designed to fit in 24GB VRAM while maintaining frontier-level intelligence for local use.
- Super VL targets specialized 'always-on' agents with high accuracy in long-context workflows.
- Both models are available under open weights, allowing for full customization and local hosting.
Related articles (1)
Open From Data Center to Desktop: NVIDIA Expands Nemotron Family With Highest Efficiency Super VL and Bolt ModelsNVIDIA Blog· 1 stories
More in Open Source
Meta Open-Sources Networking Technologies for AI Scale1 source · Oct 12Aleph Alpha Releases Kolibri Open-Weight Sovereign AI Model3 sources · Oct 5Beam Open-Weight 501b Model Debuts to Rival Proprietary Labs1 source · Oct 5Redis Creator Salvatore Sanfilippo Releases ds4 Local LLM Inference Engine1 source · Oct 2Google Releases Gemma 4 12b Encoder-Free Multimodal Model1 source · Sep 25