On October 15, 2026, NVIDIA released Nemotron 3.5 Super VL, an open 120-billion-parameter model for agents. It also launched two new edge models, Bolt 24GB and Bolt 4GB.
The Essentials
- Super VL activates 12 billion parameters per token out of its 120 billion, per NVIDIA.
- On a single GPU, Super VL delivers up to 5x higher throughput for agentic tasks compared to other open models in its size category, according to NVIDIA.
- For local hardware, the 24GB Bolt model is available now, with the 4GB Bolt model coming Oct. 23, per NVIDIA.
- Super VL ships with open weights, open datasets and training recipes. Organizations can post-train it on their own data.
NVIDIA aims the model at specialized always-on agents. Its technical blog says a single request can involve reading a document, checking information across sources, following company policy, and calling several tools.
Super VL posts NVIDIA-reported gains on agent benchmarks
Nemotron 3.5 Super VL combines reasoning, instruction following, and tool use with text, image, document, and video understanding. NVIDIA says teams can handle text and visual tasks without deploying separate text and vision-language models. On CharXiv, a chart-reasoning benchmark, the model scored 80.8%, compared with 63% for Nemotron 3 Nano Omni.
NVIDIA’s agent-harness table lists Super VL at 63.44% on Hermes Bench and 87.05% on LangChain behavioral evaluations. Nemotron 3 Ultra scores 56.33% and 84.3% on the same two tests. NVIDIA also says Super VL was distilled from Nemotron 3 Ultra.
Meet NVIDIA Nemotron 3.5 Super VL Our new open model combines reasoning and tool use with text, image, document, and video understanding for long-running AI agents. 120B total parameters with 12B active 🧵 https://t.co/BcbED05qSG
— NVIDIA AI (@NVIDIAAI) October 15, 2026
A distilled student beating its teacher on two agent tests is a strong claim. NVIDIA ran both tests itself.
Super VL delivers over 600 tokens per second on Artificial Analysis Intelligence Index agentic tasks on the Baseten endpoint, NVIDIA’s chart shows. The technical blog cites one outside check. An independent evaluation by EXO Labs placed the model on the Pareto frontier for accuracy and task completion time.
Customers tune Super VL while Bolt targets local hardware
Scale AI post-trained Nemotron 3.5 Super VL for automated vendor credit processing at Skydance Corporation. NVIDIA says inference cost for this workload will decrease by 40x. AI21 post-trained Nemotron 3.5 Super VL for AI safety and security company Alice’s threat intelligence and vulnerability analysis, improving accuracy by 40%.
Security vendors are early adopters. CrowdStrike is customizing the model for agentic blue teaming in enterprise security, per NVIDIA’s launch post.
Bolt takes the agent pitch to smaller machines. The 24GB Bolt model uses a native 2-bit quantization to fit in 24GB of memory and run alongside other applications. NVIDIA calls it the #1 open model in its class on PinchBench, and the #2 open model for BrowseComp. The 24GB checkpoint is available now on Hugging Face, with support for llama.cpp, Ollama, LM Studio, Unsloth, and vLLM.
The 4GB Bolt model is built for physical and conversational agents on GeForce RTX, Jetson, and RTX/DGX Spark.
Every speed and accuracy figure here comes from NVIDIA. With the 24GB weights public, outside testers can now check them. Next up is the 4GB model, coming Oct. 23.
