NVIDIA Expands Nemotron Family With Open Super VL and Bolt Models

Three New Open Models Cover Full Agent Deployment Range

NVIDIA today released Nemotron 3.5 Super VL, an open 120-billion-parameter model for specialized always-on agents, alongside two edge models: Nemotron 3.5 Bolt 74B-A9B 24GB, balancing intelligence and speed for local agentic workloads, and Nemotron 3.5 Bolt E4B 4GB, optimized for fast conversational agents. Super VL and Bolt 74B-A9B are available now, with Bolt E4B coming Oct. 23.

Together with Lightning, released in August for high-throughput workloads, and Ultra for hard reasoning tasks, the family covers enterprise servers down to edge devices and local hardware.

Super VL: Frontier Accuracy at One-Fifth the Size

Built with the Nemotron Coalition, Super VL delivers leading accuracy on multimodal agentic tasks, including instruction following, tool use, multistep business workflows and visual understanding, comparable with NVIDIA's largest reasoning model at roughly one-fifth the parameter count. NVIDIA says it is the fastest open agentic model in its class, with up to 5x faster throughput on the Artificial Analysis Intelligence Index leaderboard.

Unlike closed frontier models, organizations can post-train Super VL on their own data and deploy it on their own infrastructure. The model ships with open weights, datasets and training recipes, and it was trained on about 26 trillion tokens, then post-trained with roughly 3.8 million samples and reinforcement learning across dozens of environments.

Bolt Brings Agentic Intelligence to the Edge

Bolt 74B-A9B 24GB runs everyday agentic tasks locally on NVIDIA RTX and DGX Spark using native 2-bit quantization to fit in 24GB of memory. It ranks first among open models in its class on PinchBench and second on BrowseComp, letting users delegate research, writing and debugging without sending workloads to the cloud.

Bolt E4B 4GB targets GeForce RTX, Jetson and RTX/DGX Spark for local voice assistants, interactive characters and embedded interfaces.

Adoption and Availability

Scale AI, EY, Hebbia, AI21, Alice, Mercor, Ramp, Dream, Lila Sciences, R1 and others are post-training Super VL for tasks from M&A due diligence to cybersecurity, coding agents, financial documents, scientific discovery and medical claims. Atlassian, CrowdStrike, Dassault Systèmes, Perplexity, Salesforce and Synopsys are integrating it into their platforms.

The model understands images, webpages and PDFs, extracting text, reading graphs and charts so agents can move from document to data to decision without routing between systems. Super VL is available on Hugging Face, Open Router, ModelScope and as NVIDIA NIM, running on DGX Spark, DGX Station, RTX PRO, RTX and Jetson. Bolt 74B-A9B 24GB is on Hugging Face with llama.cpp, Ollama, LM Studio, Unsloth and vLLM support.

Sources: NVIDIA Blog · Nvidia Developer