We’re pleased to announce that Nebius AI Studio now hosts NVIDIA Nemotron Nano 2 VL, a compact, production-ready multimodal reasoning model engineered for real-world document intelligence and video understanding.

Built on the innovative NVIDIA hybrid Mamba-Transformer architecture, Nemotron Nano 2 VL delivers high accuracy and efficiency, making advanced vision-language intelligence accessible without the cost or latency of oversized models.

With this release, developers can now experience even more flexibility in deploying and scaling multimodal applications directly through Nebius AI Studio’s inference platform via OpenAI-compatible API.

Open, efficient and specialized AIOpen, efficient and specialized AI

Nemotron Nano 2 VL is part of the broader NVIDIA Nemotron family of open models, datasets and recipes that empower developers to build trustworthy, domain-specific AI systems.

By combining open weights, permissively licensed data and reproducible training recipes, the model provides transparency and flexibility for building enterprise-grade multimodal assistants and pipelines.

Click to expand

Model highlightsModel highlights

• High accuracy for vision and document tasks — Excellent for OCR, chart reasoning, dense image captioning and video comprehension.
• Hybrid Mamba-Transformer design — Increases throughput to process multi-image workloads faster, reducing inference cost.
• Efficient video sampling (EVS) — Processes more video at lower inference cost.
• Open and customizable — Open weights, datasets and recipes for complete transparency and model adaptation.

What developers can buildWhat developers can build

With Nemotron Nano 2 VL on Nebius AI Studio, teams can integrate multimodal reasoning into products in just a few API calls.

Developers are already using it to:

• Build document-intelligent assistants that can read dashboards, forms and diagrams with contextual understanding.
• Develop video summarization and search tools that extract scenes, captions and insights from unstructured footage.
• Automate image and media curation pipelines for ad placements and e-commerce catalogs.

Each of these use cases benefits from Nemotron Nano 2 VL’s efficiency and low latency, NVIDIA accelerated compute and Nebius’s enterprise-grade infrastructure for fast, cost-effective deployment.

Deploy with Nebius AI StudioDeploy with Nebius AI Studio

Nebius AI Studio offers a high-performance, OpenAI-compatible inference platform optimized for running Nemotron Nano 2 VL in production environments.

Developers can:

• Run inference with zero data retention.
• Scale effortlessly with usage-based pricing and dedicated endpoints with autoscaling.
• Access the model directly through the Nebius AI Studio API or the Playground.

Together, NVIDIA Nemotron Nano 2 VL and Nebius AI Studio give developers a powerful foundation for building agentic and multimodal AI systems that are open, efficient and ready for production.