Nvidia Unleashes Powerful New Open AI Models for Enterprise & PCs!
Nvidia has launched new open-weight AI models, including the Nemotron 3.5 Super VL for enterprise agents and lightweight Bolt models for local AI PCs. This move allows organizations to customize and deploy AI on their infrastructure, while also addressing consumer privacy concerns by enabling AI assistants to run on devices without cloud reliance. These models aim to foster growth in both enterprise and consumer AI markets.
Nvidia has officially launched a new suite of open-weight models, strategically designed to power always-on agents and enhance capabilities in edge computing. Central to this release is the Nemotron 3.5 Super VL, a robust large language model featuring 120 billion parameters. This model is specifically engineered for enterprise-level agents, and despite using fewer parameters than its predecessor, the Nemotron 3 Ultra, Nvidia asserts that it achieves comparable accuracy in crucial tasks like instruction-following and tool-use.
Complementing the enterprise-focused Super VL, Nvidia also introduced two lightweight Bolt models. These models are tailored for local execution on emerging AI PCs, which are powered by Nvidia's RTX Spark chip, and on workstations running the DGX Spark. These AI PCs are targeting a broad market, including consumers, programmers, and enterprise knowledge workers, and are being rolled out through key partners such as Microsoft and Dell.
Kari Briski, Nvidia’s vice president of Open Models and Software, highlighted the significance of these releases, stating that they provide organizations with an optimized, open model solution for their specific tier, eliminating the need to compromise between capability and deployability on Nvidia infrastructure. Nvidia further claims that the Super VL demonstrates superior task completion speeds compared to competing models from other major AI developers, including Anthropic's Opus 5 and OpenAI's GPT 6.1 Sol, Luna, and Astra, particularly when post-trained for specialized agentic tasks. These tasks encompass critical areas such as document intelligence, financial accounting, scientific experimentation, medical insurance claims processing, and enhancing security attack resilience.
To substantiate these performance claims, Nvidia has disclosed that several prominent companies are already leveraging Super VL, including Skydance, EY, Ramp, Crowdstrike, Perplexity, and Salesforce. This initiative marks Nvidia's continued commitment to providing open-weight models that challenge proprietary frontier labs, which are paradoxically both Nvidia's customers and investment targets. The company already boasts a comprehensive family of Nemotron models covering diverse AI applications, from multimodal intelligence to reasoning, coding, mathematics, and enterprise automation.
This strategic pivot towards open models is timely, as many rapidly growing proprietary AI labs are increasingly exploring alternative chip and AI system providers beyond Nvidia, such as AWS's Trainium and Cerebras, and even developing their own in-house chips. By emphasizing open models, Nvidia is catering to the growing demand from enterprise companies to deploy and manage their AI solutions within their own data centers. Briski underscored this by stating that, unlike closed frontier models, Super VL allows organizations to post-train it with their own data, deploy it on their proprietary infrastructure, and maintain full ownership of their creations.
Furthermore, the introduction of the smaller, local PC models signifies Nvidia's ambition to catalyze the nascent trend of consumer AI assistants. While the market is flooded with new AI assistants from startups and established tech giants, many privacy-conscious consumers are hesitant to grant such agents deep access to sensitive personal information concerning their families, health, and finances. Nvidia aims to address this by fostering a new market for AI PCs powered by its chips, enabling models and agents to run locally on devices. This approach, as articulated by Briski, allows PC users to delegate everyday projects without transmitting those workloads to the cloud, thereby enhancing privacy and control.
Delving into the Bolt models, the Nemotron 3.5 Bolt 74B-A9B 24GB is a 74-billion-parameter model specifically designed for local agentic tasks. It is engineered to strike a balance between speed and advanced reasoning capabilities, facilitating multistep reasoning, tool calling, and web browsing. The smallest variant, the Bolt E4B, is optimized for conversational agents. Nvidia has confirmed that the Super VL and Bolt 74B-A9B are currently available, with the Bolt E4B expected to be released later this month.