NVIDIA adds Nemotron 3.5 Super VL and two edge models
News
NVIDIA released Nemotron 3.5 Super VL plus Bolt 74B-A9B 24GB and Bolt E4B 4GB, expanding its open AI model family across servers and edge devices.
NVIDIA expanded its Nemotron open model family on Thursday with Nemotron 3.5 Super VL and two new edge models, extending its agentic AI lineup from data center systems to local hardware and embedded devices.
Evidence and context
In a company blog post published Oct. 15, 2026, NVIDIA said Nemotron 3.5 Super VL is an open 120-billion-parameter model aimed at specialized always-on agents, while Nemotron 3.5 Bolt 74B-A9B 24GB and Nemotron 3.5 Bolt E4B 4GB are designed for local and edge workloads. NVIDIA said Super VL and Bolt 74B-A9B are available now, with Bolt E4B due Oct. 23.
The company positioned the three models as covering a wider range of deployments, from enterprise servers to edge devices. It said Super VL and Bolt join the Nemotron family alongside Lightning and Ultra, with Lightning handling high-throughput workloads and Ultra aimed at harder reasoning tasks. The background material supplied with the release also notes that Nemotron is NVIDIA’s open AI model family and should not be confused with the earlier 2023 Nemotron-3 enterprise models.
NVIDIA said Super VL was built in collaboration with the Nemotron Coalition and is intended for multimodal agentic tasks including instruction following, tool use, multistep business workflows and visual understanding. The company said it matches the performance of its largest reasoning model at roughly one-fifth the parameter count and is up to 5x faster than the next best model in its class, based on NVIDIA’s own benchmarks and the Artificial Analysis Intelligence Index leaderboard.
For edge deployment, NVIDIA said Bolt 74B-A9B 24GB uses native 2-bit quantization to fit in 24GB of memory and run alongside other applications on systems such as NVIDIA RTX/DGX Spark. The company said it is intended for tasks including web research, writing and code debugging without sending workloads to the cloud. NVIDIA also said Bolt E4B 4GB is designed for physical and conversational agents on GeForce RTX, Jetson and RTX/DGX Spark systems.
The release also highlighted adoption by outside organizations. NVIDIA said companies including Scale AI, AfterQuery, AI21, Alice, Mercor, ModelScope, Ramp, Dream, Lila Sciences and R1 are post-training or customizing Super VL for uses ranging from finance and cybersecurity to scientific discovery and medical claims analysis. It also said software platforms including Atlassian, CrowdStrike, Dassault Systèmes, Perplexity, Salesforce and Synopsys are integrating the model into products or workflows.
NVIDIA said Super VL ships with open weights, open datasets and training recipes, and can be run on NVIDIA infrastructure with support through NVIDIA NIM and Dynamo. The company said the model is available on Hugging Face, OpenRouter, ModelScope community and as NVIDIA NIM. It also said Bolt 74B-A9B 24GB is available on Hugging Face with support for llama.cpp, Ollama, LM Studio, Unsloth and vLLM.
The announcement did not provide independent third-party verification for the performance claims in the release. Where NVIDIA cited throughput, accuracy or benchmark leadership, the figures reflect the company’s own descriptions and the stated benchmark contexts rather than a broader market consensus.
