Nvidia Corp. claims NVIDIA Nemotron 3.5 Super VL is the fastest open agentic model in its class, 5x faster than the next-best competitor based on Artificial Analysis benchmark testing against similarly sized models3,5. InclusionAI's Ling 3.0 Flash and DeepSeek's V4.1 Flash placed second and third, respectively, with what Nvidia described as significant distance between them and Super VL.
The 120B-parameter open multimodal model, released today, is built on a mixture-of-experts architecture that activates 12B parameters per forward pass4. It targets specialized enterprise agent workflows involving document reading, cross-source verification, policy compliance, and tool calling.
Nvidia Corp. said the model achieves top-tier accuracy for specialized agents with post-training. The company cited a 44% gain in medical insurance claims processing from post-training a base version of the model, a result Nvidia said put it far above Anthropic's Claude Opus 5.
Businesses can post-train the open-weight model on proprietary data and run it on their own infrastructure, which Nvidia said offers greater personalization and security.
Alongside Super VL, Nvidia Corp. released two new edge models: Nemotron 3.5 Bolt 74B-A9B 24GB, optimized to balance intelligence and speed per GB for local agentic workloads, and Nemotron 3.5 Bolt E4B 4GB, built for fast, conversational agents2.
ZDNet framed the releases as part of a broader surge in Western-made open-weight models, noting that China-based labs have held an advantage in competitive open-weight AI over the last few years but that U.S. labs are narrowing the gap.