Nvidia Nemotron 3.5: open agent models hit enterprise and PCs

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. Those tasks cover document intelligence, financial accounting, scientific experimentation, medical insurance claims and security attack resilience. Such claims are standard whenever a major tech company announces a new model, and Nvidia is labeling them as its own comparisons.

Nvidia backed the claim with a list of companies it says are using Super VL, including Skydance, EY, Ramp, Crowdstrike, Perplexity and Salesforce. These are not Nvidia's first open-weight models competing with proprietary frontier labs, which are both its customers and its investments.

The context is a shifting chip market. Fast-growing proprietary AI labs are looking beyond Nvidia to power their needs, including AWS's Trainium and Cerebras, as well as building their own home-grown chips. Nvidia's open-model push is a way to keep enterprises anchored to its infrastructure as that competition grows.

On the consumer side, Nvidia is trying to push forward the nascent trend of on-device AI assistants. Privacy-weary users may hesitate to give an agent deep access to family, health and financial information, so Nvidia argues PC users can delegate everyday projects without sending those workloads to the cloud.

For developers and Tunisian tech teams, the practical takeaway is that Super VL and Bolt 74B-A9B are available now as open-weight downloads, so organizations can post-train and self-host agents instead of relying on closed API access. Bolt E4B is expected later this month, and its small size is aimed at conversational agents on local hardware.

For buyers, Nvidia's local-agent pitch depends on the new RTX Spark AI PCs and DGX Spark workstations shipping through partners like Microsoft and Dell. Pricing and availability for these machines in Tunisia have not been announced, and neither Nvidia nor its partners have confirmed Tunisian retail channels. Nvidia has also not given a price for the models themselves, which are released as open weights rather than paid products.

Anyone waiting for local, private AI agents on a laptop should watch how the RTX Spark machines reach global markets first. If and when they arrive in Tunisia, the models will already be ready to run.