Hyperbolic is excited to host and provide Day 0 support for the latest NVIDIA Nemotron models, a new family of open models, datasets, and techniques designed to help teams build high-accuracy, high-throughput specialized agentic AI, and run it anywhere on the cloud.

We’re launching Nemotron models on Hyperbolic’s NVIDIA-powered GPU cloud and inference platform so developers can go from prototype to production with lower latency, lower cost, and full transparency into data and training processes. For teams that need to fine-tune or train their own models, our platform also provides on-demand and reserved GPUs (NVIDIA Blackwell and NVIDIA Hopper GPUs) at one of the lowest prices.

Most enterprises don’t need another giant model, they need specialized agents that understand their documents, dashboards, videos, and workflows, and that act: retrieve facts, fill forms, reconcile data, route tickets, and follow safety policies. Nemotron models provide that path with:

  • Open weights, open data, open recipes, providing auditability and trust for regulated use cases.

  • Top-tier accuracy & efficiency for vision, reasoning, and agentic tasks, engineered to do more with less time and compute.

  • Run-anywhere packaging, including NVIDIA NIM, optimized for laptops, data centers, and clouds. Ideal for Hybrid and Sovereign AI deployments.

Hyperbolic’s multi-cloud GPU infrastructure and inference stack brings Nemotron to life with on-demand capacity, production SLAs, and enterprise controls.

NVIDIA Nemotron Use Cases

  • AI assistants trained on internal data for context-aware support

  • Document and video understanding from charts, diagrams, and tables

  • RAG for intelligent search, summarization, and knowledge retrieval

  • Business process automation for workflows like reconciliations, supply chain, fraud detection, and support

  • Synthetic data generation for specialized LLM training

Nemotron Nano 2 VL (12B): Multimodal reasoning for video & documents

A compact, hybrid Transformer-Mamba VLM that delivers up to 10× higher token throughput vs prior Nano VL (preliminary) while pushing state-of-the-art accuracy across OCR, charts, video understanding, and doc intelligence. It supports 128k context, multi-image & video inputs, and outputs text.

The model has leading accuracy on OCRBench v2 and 73.2 on average score across the following benchmarks: MMMU, MathVista, AI2D, OCRBench, OCRBench-v2, OCR-Reasoning, ChartQA, DocVQA, and Video-MME

Use Cases

  • AI assistant with document intelligence for customer service (dashboards, screenshots, docs), IT, finance, insurance, healthcare forms

  • AI assistant for video understanding, video and image curation for training VLMs

  • Ingestion of image or videos - dense caption of image/videos

  • RAG, Multi-modal agentic apps and services

On Hyperbolic: accessible via API endpoints for seamless integration into apps or pipelines.

Nemotron Parse 1.1: Lightweight parsing from complex docs

A lightweight ~1B VLM built to extract structured info from PDFs, contracts, research reports, statements, charts, and diagrams. Available soon on Hyperbolic.

Get Started Now on Hyperbolic

Check out our inference page, with the latest NVIDIA Nemotron models here.

For on-demand GPUs (H100s and H200s), check it out here.

About Hyperbolic

Hyperbolic is the on-demand AI cloud made for developers. We provide fast, affordable access to compute, inference, and AI services. Over 195,000 developers use Hyperbolic to train, fine-tune, and deploy models at scale.

Our platform has quickly become a favorite among AI researchers, including those like Andrej Karpathy. We collaborate with teams at Hugging Face, Vercel, Quora, Chatbot Arena, LMSYS, OpenRouter, Black Forest Labs, Stanford, Berkeley, and beyond.

Founded by AI researchers from UC Berkeley and the University of Washington, Hyperbolic is built for the next wave of AI innovation—open, accessible, and developer-first.

Website | X | Discord | LinkedIn | YouTube | GitHub | Documentation