AI news ·
Nvidia releases Nemotron 3.5 Super VL and two Bolt models for document analysis and local agent tasks
NVIDIA released Nemotron 3.5 Super VL, a 120-billion-parameter open model combining language and vision that enterprises can post-train on proprietary documents. After customization, medical-claims accuracy jumped from roughly 28% to 72%, outpacing the Claude Opus 5 score of about 38% on that task.
NVIDIA released Nemotron 3.5 Super VL, a 120-billion-parameter open model that combines language and visual understanding, alongside two smaller Bolt models for local agents. The launch gives enterprise developers a model they can post-train on proprietary documents and procedures, then deploy on infrastructure matching their security and data-residency requirements.
Super VL handles PDFs, charts, tables, images and webpages while following instructions and using tools. Bolt 74B-A9B 24GB targets local coding and research tasks, while Bolt E4B 4GB, due October 23, is aimed at conversational and device-based agents.
Where visual understanding fits in the Nemotron family
Super VL is not NVIDIA's first model to read visual material. Nemotron Nano 2 VL already offered document and image analysis, and April's Nemotron 3 Nano Omni combined text, images, audio and video in a 30-billion-parameter model. The change is where visual understanding sits in the broader lineup. NVIDIA released a 120-billion-parameter Nemotron 3 Super model for agentic reasoning in March, followed by the larger Nemotron 3 Ultra. August's Nemotron 3.5 Lightning targets repeated, high-volume steps in longer agent workflows. Super VL brings document and image understanding into the larger Super tier, while Bolt extends local deployment options.
That combination could simplify some agent designs. A system reviewing an insurance claim might need to read scanned records, extract structured information, check it against a policy and decide what to flag. Handling text and visual material in one model can reduce the need to pass information between separate reading systems. Developers still have to test whether the model extracts the right facts and follows the relevant rules.
What partners report after customization
NVIDIA says dozens of organizations are post-training Super VL for their own tasks. Its announcement names work by Scale AI for Skydance's vendor-credit finance agent, EY's merger-and-acquisition due-diligence agent, AfterQuery for Hebbia's credit-agreement analysis, and AI21 and Alice for cybersecurity. Platform companies integrating it include Atlassian, CrowdStrike, Dassault Systèmes, Perplexity, Salesforce and Synopsys.
An NVIDIA chart shows five separate evaluations: document intelligence, security attack resilience, financial receipt extraction, scientific experimentation and medical insurance claims. The improvements after post-training range from 9 to 44% across the five tasks. In each comparison shown, the resulting specialized model scores above the named Claude or GPT model tested on that task. The medical-claims result shows the largest gain: accuracy rises from roughly 28% to 72% after post-training, compared with about 38% for Claude Opus 5. Security attack resilience rises from roughly 35% to 76%, compared with about 66% for GPT-6.1 Sol.
These are different customized versions on different specialized tasks, not a claim that the base Super VL model beats those models generally. NVIDIA has not supplied the task definitions, sample sizes or testing protocol, so a buyer should test the same approach against its own work.
Speed and the local-agent Bolt models
NVIDIA describes Super VL as up to five times faster than a comparable open model. Its speed chart plots the Artificial Analysis Intelligence Index against output tokens per second with 100,000 input tokens. Super VL appears near 625 output tokens per second at an index score of about 26, measured on four Blackwell B200 GPUs. The fivefold comparison arrow begins at Inkling Small, near 125 tokens per second and a similar index score. The result makes Super VL unusually fast at that score, though several open models score higher on the composite index, including GLM-5.3-Flash at 42 and Qwen3.8-Flash-Next at 40.
Output speed matters when an agent makes many calls over a long job, but it is distinct from total completion time. Reading inputs, calling outside tools, checking results and correcting mistakes all contribute. The chart establishes a generation-speed comparison under its stated test conditions; it does not show a fivefold reduction in end-to-end workflow time.
Bolt 74B-A9B has 74 billion total parameters but activates 9 billion at a time. Native 2-bit compression lets it fit in 24GB of memory and run alongside other applications on RTX Spark and DGX Spark systems. It targets web research, writing and debugging code through local agent tools. The model is available on Hugging Face with support for llama.cpp, Ollama, LM Studio and vLLM. Bolt E4B takes the local approach down to 4GB for voice assistants, interactive characters and embedded interfaces, targeting GeForce RTX, Jetson and RTX/DGX Spark when it arrives October 23.
Access, licensing and production costs
Super VL's open weights are available on Hugging Face, with OpenRouter, ModelScope and NVIDIA NIM as additional launch access routes. Bolt 74B-A9B is also on Hugging Face. The models use NVIDIA's OpenMDW license, which permits commercial use, fine-tuning and redistribution. NVIDIA is releasing training recipes and datasets, and provides customization resources through NeMo and a technical architecture guide.
Developers can use a rate-limited NVIDIA-hosted API at no charge for development. NVIDIA partners offer hosted APIs at prices set by each provider, including through OpenRouter. Organizations can download NIM microservices from NVIDIA's NGC catalog. For production, NVIDIA quotes an NVIDIA AI Enterprise license of $4,500 per GPU per year - a software-license figure, not the total cost of GPUs, hosting and operations.
The launch follows NVIDIA's agreement to acquire Hugging Face for roughly $12.93 billion. The deal has not closed; NVIDIA expects completion in the first half of 2027, subject to regulatory approval. The company has committed to keeping Hugging Face open to other model makers and chip vendors.
Why this matters for IT and development teams
Super VL offers a concrete test case: whether an open model you can adapt and deploy under your own control reads your documents and follows your procedures accurately enough to justify its cost. The partner examples and speed chart show what to investigate, not what every deployment will achieve. Bolt provides a separate option for workloads you want to keep local. NVIDIA now provides the models, access routes and a production software price to begin those evaluations. For teams building document-heavy agents - insurance claims, credit analysis, M&A due diligence - the ability to handle text and visual material in one model may reduce pipeline complexity, but only if extraction accuracy holds up on your actual documents.