Open From Data Center to Desktop: NVIDIA Expands Nemotron Family With Highest Efficiency Super VL and Bolt Models

NVIDIA today announced the release of Nemotron 3.5 Super VL, a 120‑billion‑parameter open model that delivers the fastest task completion and highest accuracy for specialized always‑on agents. Joining Super VL are two new edge models: Nemotron 3.5 Bolt 74B‑A9B 24GB, which balances 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. Super VL and Bolt 74B‑A9B are available immediately, with Bolt E4B slated for release on Oct. 23. The three models together cover the full spectrum of agent deployments, from large enterprise servers to edge devices and local hardware. Super VL is built in collaboration with the Nemotron Coalition and achieves leading accuracy on multimodal agentic tasks such as instruction following, tool use, multistep business workflows, and visual understanding, comparable to NVIDIA’s largest reasoning model at roughly one‑fifth its parameter count. It is five times faster than the next best open agentic model, enabling enterprise agents to complete high‑volume tasks more quickly across extended workflows. Unlike closed frontier models, organizations can post‑train Super VL on their own data, deploy it on their own infrastructure, and maintain oversight of their builds. Bolt 74B‑A9B 24GB is designed to run everyday agentic tasks locally on NVIDIA RTX/DGX Spark, using native 2‑bit quantization to fit within 24 GB of memory and run alongside other applications. Relative to other models with a similar memory footprint, it offers a good balance of intelligence and speed, and it is the #1 open model on PinchBench and the #2 open model for BrowseComp. Bolt E4B 4GB is built for physical and conversational agents on GeForce RTX, Jetson, and RTX/DGX Spark, combining natural conversation with multi‑turn agentic capabilities to power fast local voice assistants, interactive characters, and embedded interfaces. NVIDIA is also releasing open datasets on Hugging Face covering agentic tool use, finance, instruction following, long context, fairness, and vision‑language understanding, along with reinforcement learning environments in NeMo Gym. The models are packaged as NVIDIA NIM microservices with performance‑optimized, validated configurations and enterprise‑grade support, refreshed every two weeks with security patches and performance improvements. Nemotron 3.5 Super VL runs across NVIDIA’s full hardware stack, including DGX Spark, DGX Station, RTX PRO, RTX systems, NVIDIA Jetson, and data center and cloud environments. Bolt 74B‑A9B 24GB is available on Hugging Face with support for llama.cpp, Ollama, LM Studio, Unsloth, and vLLM, and it is built for NVIDIA RTX Spark laptops arriving Oct. 16. The release provides organizations with open, optimized models for their tier, eliminating the trade‑off between capability and deployability.