N環球 
Nvidia launches Nemotron 3.5 open models for enterprise agents and local AI PCs
Nvidia's new 120-billion-parameter Nemotron 3.5 Super VL and two lightweight Bolt models ship with open weights, datasets and training recipes, targeting always-on agents in the enterprise and on local AI PCs.
Why it matters
Nvidia is pulling the AI-agent battlefield off the cloud and into the enterprise and the personal computer: the whole Nemotron 3.5 family ships with open weights, datasets and training recipes, so companies can fine-tune and deploy on their own infrastructure with their own data — a key step as firms demand data sovereignty and privacy.
Full report
On October 15, Nvidia released the Nemotron 3.5 family of open-weight models built for always-on agents and edge computing. The flagship, Nemotron 3.5 Super VL, packs 120 billion parameters in a mixture-of-experts architecture that activates only about 12 billion parameters per token. Nvidia says it matches the accuracy of its larger reasoning model, Nemotron 3 Ultra, on instruction-following and tool-use agent tasks.
The company also claims Super VL completes post-trained agentic tasks faster than comparable models from Anthropic and OpenAI (Opus 5, GPT 6.1 Sol) and tops its class on the Artificial Analysis leaderboard with up to 5x the throughput — the kind of claim that accompanies every vendor launch and needs independent verification. Nvidia lists early enterprise users including Skydance, EY, Ramp, Crowdstrike, Perplexity and Salesforce; the model is available on Hugging Face, OpenRouter, ModelScope and NVIDIA NIM.
Two lightweight Bolt models target local compute: Nemotron 3.5 Bolt 74B-A9B uses native 2-bit quantization to fit in 24GB of memory and runs agent tasks (multi-step reasoning, tool calling, web browsing) locally on new AI PCs with Nvidia RTX Spark chips and DGX Spark workstations, now shipping through partners including Microsoft and Dell. The smaller Bolt E4B (4GB), aimed at voice assistants and conversational agents, follows later this month.
Nvidia's vice president of Open Models and Software, Kari Briski, said: "Any organization that wants to run agents on Nvidia infrastructure now has an open, optimized model for their tier without having to choose between capability and deployability."
💬 Comments
No comments yet. Be the first!