Nvidia's new open agent models claim 5x the speed of the next best in class — and one is built to run agents locally in 24GB

“Nvidia says its new open agent model is 5x faster than the next best in its class — and a sibling model is built to run agents locally in 24GB of memory.”

Usable claims

• On October 15, 2026 (Thursday, calendar-verified), NVIDIA released Nemotron 3.5 Super VL — per the company's blog post by Kari Briski, 'an open 120-billion-parameter model that delivers the fastest task completion and highest accuracy for specialized always-on agents' — together with two 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, optimized for intelligence per GB for fast conversational agents'. Availability, per the post: 'Super VL and Bolt 74B-A9B are available now, with Bolt E4B coming Oct. 23.' Size framing, per the post: accuracy 'comparable with NVIDIA's largest reasoning model at roughly one-fifth the parameter count'. Speed claims are the company's own and keep their shapes: 'As the fastest open agentic model in its class — 5x faster than the next best'; and 'Nemotron 3.5 Super VL, on Artificial Analysis Intelligence Index leadboard, achieves up to 5x faster throughput compared to other models in its class.' (the post's own spelling 'leadboard' and its 'up to' hedge). Openness, per the post: 'Unlike closed frontier models, organizations can post-train Super VL on their own data, deploy it on their own infrastructure and oversee what they build.' Local-agent layer, per the post: Bolt 74B-A9B 24GB is 'designed to run everyday agentic tasks locally on NVIDIA RTX/DGX Spark, using a native 2-bit quantization to fit in 24GB of memory'; 'It is the #1 open model in its class on PinchBench, and the #2 open model for BrowseComp.'; and it is 'Built for NVIDIA RTX Spark laptops arriving Oct. 16'. Family context, per the post: 'Super VL and Bolt join the Nemotron family alongside Lightning and Ultra.' Training scale, per the post: 'about 26 trillion tokens' across pretraining and continued pretraining, 'roughly 3.8 million samples' of post-training, and reinforcement learning 'across dozens of environments'; and with NVIDIA Dynamo, 'organizations can serve up to 8x more users with 85% lower latency'. Per TechCrunch (Julie Bort, 8:37 AM PDT October 19): 'Nvidia on Thursday launched new open-weight models designed for always-on agents and edge computing.' Super VL is 'geared for agents used by the enterprise'; the Bolt models 'are intended to run locally on the new AI PCs (running the Nvidia RTX Spark chip) and workstations running the DGX Spark', which are 'being released right now into the market through partners like Microsoft and Dell'; and 'Nvidia claims that the Super VL completes tasks faster than competing models from Anthropic and OpenAI (Opus 5, GPT 6.1 Sol, Luna, Astra) when post-trained on specific agentic tasks.' The task list, per the article: 'document intelligence, financial accounting, scientific experimentation, medical insurance claims, and security attack resilience'.