Nvidia introduced a new suite of open-weight artificial intelligence models on Thursday, targeting both enterprise data centers and personal computers. The release includes the Nemotron 3.5 Super VL for large-scale organizational use and two lightweight Bolt models designed to run locally on new AI-enabled hardware.
The Super VL model contains 120 billion parameters and is built specifically for always-on agents used in business environments. Nvidia states that despite having fewer parameters than its Nemotron 3 Ultra model, the Super VL delivers comparable accuracy in instruction following and tool-use tasks. The company claims the model completes specific agentic tasks faster than competing proprietary models from Anthropic and OpenAI when post-trained on specialized workloads.
Nemotron 3.5 Super VL for Enterprise Infrastructure
Nvidia positioned the Super VL as a solution for organizations seeking to maintain control over their data and infrastructure. The model supports tasks such as document intelligence, financial accounting, scientific experimentation, medical insurance claims processing, and security attack resilience.
Kari Briski, Nvidia’s vice president of Open Models and Software, emphasized the strategic advantage of open-weight models in a blog post. She noted that organizations can post-train the Super VL on their own data and deploy it on their own infrastructure, allowing them to own the resulting capabilities without relying on closed frontier models.
Several major companies are reportedly already using the Super VL. Nvidia listed Skydance, EY, Ramp, CrowdStrike, Perplexity, and Salesforce as early adopters. This move aligns with a broader trend of enterprises wanting to run their own models in-house rather than depending entirely on external cloud providers.
Bolt Models for Local AI Computing
At the other end of the spectrum, Nvidia launched two Bolt models intended for local execution on consumer and professional hardware. These models are designed to run on AI PCs powered by the Nvidia RTX Spark chip and workstations using the DGX Spark system. Partners including Microsoft and Dell are currently releasing devices equipped with these chips.
The Nemotron 3.5 Bolt 74B-A9B 24GB is a 74-billion parameter model aimed at local agentic tasks. It is designed to balance speed with reasoning capabilities to handle multistep reasoning, tool calling, and web browsing. The smaller Bolt E4B model is targeted at conversational agents.
Nvidia hopes these local models will address privacy concerns among consumers who are hesitant to share sensitive personal information with cloud-based AI assistants. By processing workloads locally, users can delegate everyday projects without sending data to external servers.
The Super VL and Bolt 74B-A9B models are available immediately. The Bolt E4B model is scheduled for release later this month. These releases add to Nvidia’s existing Nemotron family, which includes models for multimodal AI, reasoning, coding, and math.
Source: TechCrunch
