Nvidia on Thursday launched new open-weight models designed for always-on agents and edge computing. TechCrunch reports that the release includes the large Nemotron 3.5 Super VL for corporate use and two lightweight Bolt models built to run locally on hardware such as new AI PCs and workstations.
Key takeaways
- Nvidia introduced the 120-billion-parameter Nemotron 3.5 Super VL model for enterprise agent tasks.
- Two new Bolt models target local execution on workstations and AI PCs powered by the Nvidia RTX Spark chip.
- Companies like Salesforce, EY, and Perplexity are already using the Super VL model.
- The smaller Bolt E4B conversational model arrives later this month, while Super VL and Bolt 74B-A9B are available now.
What are the new Open models from Nvidia?
The headline release is the Nemotron 3.5 Super VL, a 120-billion-parameter model built for enterprise agents. While it packs fewer parameters than the Nemotron 3 Ultra model, Nvidia states it matches that older version in instruction-following and tool-use accuracy. The company claims Super VL finishes tasks faster than competing systems from OpenAI and Anthropic when post-trained for specific agentic duties like financial accounting and document intelligence.
To back up these performance claims, Nvidia points to heavy-hitter customers already testing Super VL. These organizations include Salesforce, Perplexity, Crowdstrike, Ramp, EY, and Skydance. Kari Briski, Nvidia’s vice president of Open Models and Software, noted in a blog post that corporations can post-train Super VL on private data and own what they build.
How do the Bolt models work on local hardware?
For consumer hardware, Nvidia debuted two lightweight Bolt models aimed at local execution on AI PCs running the Nvidia RTX Spark chip and DGX Spark workstations. These machines roll out through partners like Microsoft and Dell for consumers, programmers, and enterprise knowledge workers. Users can delegate everyday tasks without sending workloads to the cloud, addressing privacy concerns regarding family, health, and financial data.
The larger Nemotron 3.5 Bolt 74B-A9B 24GB features 74 billion parameters for multistep reasoning, web browsing, and tool calling. Meanwhile, the smaller Bolt E4B focuses on conversational agent duties.
What it means for enterprise users
As proprietary AI labs increasingly rely on diverse hardware makers beyond Nvidia, the GPU giant focuses heavily on companies wanting to run custom Open models inside their own data centers. Organizations now have optimized options for their specific tiers. Briski explained that any group wishing to run agents on Nvidia infrastructure can use an open model without compromising between deployability and capability.
Frequently asked questions
The Nemotron 3.5 Super VL is a large, 120-billion-parameter open-weight model created by Nvidia for enterprise agents. It delivers high accuracy in instruction-following and tool-use tasks while completing jobs faster than certain competing models when post-trained.
Nvidia reports that heavy-hitter companies already using the Super VL model include Salesforce, Perplexity, Crowdstrike, Ramp, EY, and Skydance.
The Bolt models are lightweight Open models designed to run locally on AI PCs and workstations. They include the 74-billion-parameter Bolt 74B-A9B for local agentic tasks and the smaller Bolt E4B for conversational agents.
Nvidia states that the Super VL and the Bolt 74B-A9B models are available right now. The smaller Bolt E4B conversational model is scheduled to arrive later this month.
Running AI models locally on PCs allows privacy-weary consumers and workers to delegate everyday projects and tasks without sending sensitive family, health, or financial workloads to the cloud.
Based on reporting by TechCrunch. Read the original article
