Nvidia launches Nemotron open models for enterprise and PC agents
Nvidia introduced open-weight models for enterprise agents and edge computing: Nemotron 3.5 Super VL and Bolt. Super VL, with 120B parameters (12B active), targets enterprise agentic tasks, scores 26 on the Artificial Analysis Intelligence Index (double its predecessor's 13), features a 1-million-token context window, and costs $0.23 per million input tokens via API. It processes text, images, web pages, and PDFs. Bolt, a 74B-parameter (9B active) lightweight model, runs locally on AI PCs and workstations in 24GB memory via 2-bit quantization. Both are available now under the permissive OpenMDW-1.1 license, with Bolt E4B later this month. Nvidia emphasizes post-training on own data and deployment on own infrastructure. Partners include Salesforce, Atlassian, CrowdStrike, Perplexity, Skydance, EY, Ramp, Synopsys, and Dassault Systèmes. Models are on Hugging Face, OpenRouter, and as Nvidia microservices. Nvidia is in talks to acquire Reflection AI.