Nvidia on Oct 15, 2026 released Nemotron 3.5 Super VL, an open 120-billion-parameter model it says delivers the fastest task completion and highest accuracy for specialized always-on agents, claiming up to 5x faster throughput than others in its class on the Artificial Analysis Intelligence Index. Two edge models join it: Nemotron 3.5 Bolt 74B-A9B (24GB, for local agentic work on RTX/DGX Spark, using native 2-bit quantization) and Bolt E4B (4GB, for conversational agents on GeForce RTX, Jetson and Spark). Super VL and Bolt 74B are available now; Bolt E4B arrives Oct. 23. Nvidia says organizations can post-train the open models on their own data and run them on their own infrastructure, and lists Scale AI, Salesforce, CrowdStrike, Perplexity and Atlassian among adopters.
Our takeNvidia's openness is really a distribution play: if enterprises post-train and self-host Nemotron, they stay on Nvidia silicon from data center to laptop. The "one-fifth the size" framing and the 5x throughput claim come straight from the vendor, so treat them as marketing until independent testing checks the numbers. The open question is whether an open model can hold up against closed frontier agents on messy, multi-day workflows — which is exactly the workload Nvidia is aiming at.