Nvidia released an open-weight AI model called Nemotron 3.5 Super VL, a 120-billion-parameter system the company says is built for the growing market for agentic AI tools used by enterprises. The model is available as of the announcement. Nvidia claims it delivers the fastest task completion and highest accuracy for specialized, always-on agents, according to its release.
Nvidia said Super VL is the fastest open agentic model in its class, five times faster than the next best option, allowing enterprise agents to finish high-volume work across workflows that can span days or weeks. That claim is based on Artificial Analysis benchmark testing against similarly sized competitors. InclusionAI's Ling 3.0 Flash placed second and DeepSeek's V4.1 Flash placed third, with a significant gap between them and Super VL.
The company also said Super VL reaches top-tier accuracy for specialized agents after post-training. Nvidia pointed to one case in which post-training a base version of the model produced a 44% gain in medical insurance claims, putting it far above Claude Opus 5. The base model did not beat proprietary frontier models on its own, but Nvidia says enterprises can tailor it to their own needs for highly specific gains.
Nvidia argues that open models offer businesses advantages over proprietary ones because companies can post-train them on their own data and run them on their own infrastructure, providing more personalization and security.
Two new edge models accompany Super VL: Nemotron 3.5 Bolt 74B-A9B 24GB and Nemotron 3.5 Bolt E4B 4GB. Edge AI refers to models that run locally and off-grid. Nvidia said the 24GB Bolt is designed for local agentic use, while the 4GB Bolt is for fast, conversational agents.
Both models provide the highest intelligence per gigabyte of memory among Nvidia's models, the company said. The 24GB version achieves the best balance of capability and speed per gigabyte, while the 4GB version offers the best intelligence per gigabyte for uses Nvidia identified as voice and device agents and interactive characters, among others.
The release comes as Western open models push back against the lead Chinese labs have held in competitive open-weight AI. French lab Mistral recently launched its Le Chonk model, which hit certain state-of-the-art benchmarks, and Reflection AI, which Nvidia has invested in and is in talks to acquire, released its Beam model last week. The industry has also warmed to open-weight models after the OpenAI Hugging Face breach undercut the idea that only open models carry security risks. If US-made open models prove competitive with Chinese ones, that could help adoption in areas covered by cybersecurity guidance against foreign technology.
More AI news from TechManNews.
