Nvidia on Thursday launched three new open-weight models, the latest additions to its Nvidia open models range, built for always-on AI agents and for edge computing, which means processing data on devices close to where it is created. The launch has two audiences: large companies that want to run agents inside their own data centers, and people who use the new AI PCs.
Two models for different hardware
The largest model is Nemotron 3.5 Super VL, a 120-billion-parameter model aimed at agents used by enterprises. Parameters are the internal settings that shape how a model behaves, so a higher count generally means a larger model. Nvidia said Super VL uses fewer parameters than its Nemotron 3 Ultra model but can deliver similar accuracy on instruction-following and tool-use tasks.
The two lightweight Bolt models are meant to run locally on AI PCs built around the Nvidia RTX Spark chip, and on workstations running DGX Spark. Nemotron 3.5 Bolt 74B-A9B 24GB has 74 billion parameters and is designed for local agent work such as multistep reasoning, tool calling and web browsing. The smallest model, Bolt E4B, is aimed at conversational agents and arrives later this month. Super VL and Bolt 74B-A9B are available now.
Enterprise AI in data centers and Nvidia's claims
Nvidia said Super VL completes tasks faster than competing models from Anthropic and OpenAI when it is post-trained on specific agentic tasks. Post-training means further training a model on particular data for a particular job. The listed tasks include document intelligence, financial accounting, scientific experimentation, medical insurance claims and security attack resilience. These are the company's own comparisons, not independent results.
Nvidia named Skydance, EY, Ramp, CrowdStrike, Perplexity and Salesforce as companies using Super VL. Kari Briski, Nvidia's vice president of Open Models and Software, wrote in the announcement that organizations can post-train the model on their own data and deploy it on their own infrastructure, so they "own what they build." The approach contrasts with closed frontier models, which companies cannot modify in the same way.
The move comes as proprietary AI labs, which are also Nvidia's customers and investments, look to other chip and system makers for their own needs. Nvidia's Nemotron family already covers multimodal AI, reasoning, coding, math and enterprise automation. The new models extend that family toward agents that run on Nvidia hardware.
Local agents on AI PCs
The Bolt models are the company's bet on consumer AI assistants that stay on the device. Briski wrote that PC users "can delegate everyday projects" without sending that work to the cloud. This matters to privacy-conscious users who may be reluctant to give assistants access to family, health or financial information.
The PCs are being released now through partners including Microsoft and Dell, and they are aimed at consumers, programmers and enterprise knowledge workers. Nvidia hopes the machines will create a new market for PCs powered by its chips. The launch did not come with prices.
Aasan alfaaz mein
Nvidia ne teen naye open models launch kiye hain. Inmein se 120 billion parameter wala Super VL company ke bade agents ke liye hai aur data center mein chalta hai. Do chhote Bolt models AI PC aur workstation par locally chal sakte hain, aur Super VL aur Bolt 74B-A9B abhi available hain. Aam user ke liye iska matlab hai ke AI assistant apne device par kaam kar sakta hai, lekin is launch mein qeemat nahi batayi gayi.
