At the other end sit two lightweight Bolt models. Nemotron 3.5 Bolt 74B-A9B 24GB, a 74-billion-parameter model, is designed to run locally on AI PCs using Nvidia's RTX Spark chip and on DGX Spark workstations, balancing speed with multistep reasoning, tool calling and web browsing. The smaller Bolt E4B targets conversational agents. Super VL and Bolt 74B-A9B are available now, with Bolt E4B due later this month.

Nvidia claims Super VL completes tasks faster than competing models from Anthropic and OpenAI, including Opus 5, GPT 6.1 Sol, Luna and Astra, when post-trained on specific agentic tasks such as document intelligence, financial accounting and medical insurance claims. Such vendor benchmarks are standard fare at model launches, but Nvidia has also named companies it says are using Super VL, including Skydance, EY, Ramp, Crowdstrike, Perplexity and Salesforce.

The open-weight push also has a strategic edge. TechCrunch notes that fast-growing proprietary AI labs are increasingly looking beyond Nvidia to chips from AWS, Cerebras and their own in-house designs, so Nvidia is leaning into the demand from enterprises that want to run open models in their own data centres. "Unlike closed frontier models, organizations can post-train Super VL on their own data, deploy it on their own infrastructure and own what they build," Kari Briski, Nvidia's vice president of Open Models and Software, wrote in the announcement blog post.

On the consumer side, the Bolt models are meant to advance the nascent trend of local AI assistants, letting PC users hand over everyday projects without sending those workloads to the cloud, which may appeal to people wary of sharing family, health or financial information with cloud-based agents.