Breaking

NVIDIA Releases Nemotron 3.5 Super VL: 120B Open Vision-Language Model for Specialized Agents

October 21, 2026 at 00:26 EDT

Foundation Models
AI Agents

NVIDIA has released Nemotron 3.5 Super VL, a 120-billion-parameter open-weights vision-language model designed to power specialized AI agents with text, image, document, and video reasoning capabilities. The model, released on October 15, 2026, combines multimodal understanding with tool use and structured reasoning in a single system, challenging the dominance of proprietary vision-language models in enterprise AI applications.

October 15, 2026 · NVIDIA AI

NVIDIA Nemotron 3.5 Super VL: Open Vision-Language Model at Scale

A 120-billion-parameter mixture-of-experts model combining text, image, document, and video understanding for specialized enterprise AI agents.

Performance at a Glance

63.44%

Hermes Bench Score

87.05%

LangChain Evaluation

80.8%

Chart Reasoning

5×

Agentic Task Throughput

Architecture & Deployment

Model Specifications

120B total parameters
12B active per token
1M token context
Hybrid Mamba-2 + MoE

Availability

Open weights (Hugging Face)
BF16 & NVFP4 quantized
OpenMDW-1.1 license
$0.10 / 1M input tokens

Key Capability: Fine-tuning on domain-specific tasks shows rapid adaptation—chart-reading accuracy improved from 17% to 78% after 3.3 hours of training on four H200 GPUs, suitable for document intelligence and custom enterprise workflows.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.