Nvidia launches Nemotron open models for agents, PCs — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Nvidia launched the 120-billion-parameter Nemotron 3.5 Super VL, an open-weight model for enterprise agents, claiming it completes tasks faster than competing models from Anthropic and OpenAI on agentic workloads like document intelligence, financial accounting, medical claims, and security attack resilience.
- Nvidia also released two Bolt models — the 74B-parameter Bolt 74B-A9B for local agentic tasks and the smaller Bolt E4B for conversational agents — designed to run on AI PCs powered by its RTX Spark chip and DGX Spark workstations, shipping now through partners Microsoft and Dell.
- Nvidia named six enterprise customers already using Super VL: EY, Skydance, Ramp, Crowdstrike, and Salesforce, plus Perplexity.
- Nvidia VP Kari Briski framed the open-weight approach as letting organizations post-train on their own data and deploy on their own infrastructure — a contrast to closed frontier models that lock customers into run them.
- Super VL and Bolt 74B-A9B are available now; Bolt E4B arrives later this month.
- Anthropic and OpenAI, both Nvidia customers and Nemotron competitors, are diversifying beyond Nvidia to AWS Trainium, Cerebras, and homegrown silicon — the backdrop against which Nvidia is leaning into open-weight, self-hosted models to stay central to run them.
Why it matters: Nvidia's biggest AI lab customers are diversifying to AWS Trainium, Cerebras, and homegrown chips — so Nvidia is countering with a model-plus-chip stack: open-weight Nemotron models enterprises can self-host (with EY, Crowdstrike, Perplexity and three others named) plus RTX Spark PCs running agents locally for privacy-wary consumers. Owning the model layer keeps Nvidia central as the chip market fragments.
Ask SkimNews
