Nvidia releases Nemotron 3.5 Super VL and Bolt edge models
Nvidia has released three new open-weight artificial intelligence models under its Nemotron 3.5 family, targeting agent deployments across enterprise servers, edge devices, and local hardware.
The flagship model, Nemotron 3.5 Super VL, is a 120-billion-parameter vision-language model built in collaboration with the Nemotron Coalition. According to Nvidia, it delivers up to 5x faster throughput compared to other models in its class, based on the Artificial Analysis Intelligence Index leaderboard. The model supports multimodal tasks including optical character recognition, table and chart reading, tool use, and multistep workflows.
Two edge-focused models accompany the release: Nemotron 3.5 Bolt 74B-A9B 24GB, designed for local agentic workloads on Nvidia RTX and DGX Spark systems using native 2-bit quantization to fit within 24GB of memory, and Nemotron 3.5 Bolt E4B 4GB, targeting conversational and physical agents on GeForce RTX, Jetson, and RTX/DGX Spark hardware. Super VL and Bolt 74B-A9B are available now; Bolt E4B is scheduled for release Oct. 23.
Nvidia states the model was trained on approximately 26 trillion tokens across pretraining stages and post-trained using roughly 3.8 million samples with reinforcement learning. Open weights, datasets, and training recipes are being released on Hugging Face alongside reinforcement learning environments through NeMo Gym.
Several organizations have reportedly post-trained Super VL for specialized applications. These include Scale AI for Skydance and EY, CrowdStrike for enterprise security, Salesforce for government workflow automation, Atlassian for agentic search, Perplexity for search tasks, and Dassault Systèmes for engineering design. Additional adopters include Ramp, Mercor, and TrainLoop, among others.
Nvidia packages the model as a production microservice through Nvidia NIM and says deployments running on Nvidia Dynamo can serve up to 8x more users with 85% lower latency. The model runs across Nvidia's hardware stack, including DGX Spark, DGX Station, RTX PRO, Jetson, and cloud environments.