Nvidia Opens Nemotron 3.5 Super VL, a 120B Multimodal Model for Always-On AI Agents, Plus Local Bolt Models

Nvidia on Thursday, Oct. 15, 2026 released Nemotron 3.5 Super VL, an open 120-billion-parameter multimodal model aimed at specialized, always-on AI agents that need to read documents, charts and web pages while using tools. The company also launched Nemotron 3.5 Bolt 74B-A9B 24GB for local agent work and said a smaller Bolt E4B 4GB model arrives Oct. 23. Nvidia's own blog and separate coverage from VentureBeat and Wccftech all confirm the lineup.

On X, Nvidia AI described Super VL as 120B total parameters with 12B active, combining reasoning and tool use with text, image, document and video understanding. The company blog calls it the fastest open agentic model in its class and says it is up to 5x faster than the next-best model in that comparison on Artificial Analysis throughput charts.

What ships today

• Super VL: Open weights, datasets and training recipes. Nvidia says it was trained on about 26 trillion tokens and post-trained with roughly 3.8 million samples. It targets OCR, tables, charts and structured info in PDFs and web pages so agents do not have to split vision and text across separate models.
• Bolt 74B-A9B 24GB: A mixture-of-experts edge model with 74 billion total parameters that activates 9 billion at a time, using native 2-bit quantization to fit in 24GB for local research, coding and tool use on RTX/DGX Spark and related hardware.
• Bolt E4B 4GB: Due Oct. 23 for conversational and embedded agents on GeForce RTX, Jetson and RTX/DGX Spark.

Nvidia says dozens of organizations are already post-training Super VL for document intelligence, cybersecurity, finance, science and medical claims, and names platform integrators including Atlassian, CrowdStrike, Perplexity, Salesforce and Synopsys. Weights are listed on Hugging Face, OpenRouter, ModelScope and as Nvidia NIM microservices.

What it means for Tesla

This is an Nvidia software story, not a Tesla product launch. The Tesla angle is the AI compute race Tesla and xAI are already in: Tesla is pursuing its own inference chips and Terafab talks with TSMC, while much of the industry still trains and serves agents on Nvidia GPUs. Open, customizable agent models that run on Nvidia's stack raise the bar for what rivals can ship without a closed frontier API — the same category Tesla fans watch when comparing Grok, Optimus software and in-car assistants. For Huang's separate safety comments the same day, see our Jensen Huang AI safety story.