NVIDIA Nemotron 3 Super now available on Nebius Token Factory

NVIDIA Nemotron 3 Super is now available on Nebius Token Factory.

Nemotron 3 Super is a 120B parameter hybrid MoE model optimized for multi-agent applications and complex reasoning workflows. With 12B active parameters per inference step and up to 1M token context length, it is designed for long-horizon planning, tool calling and high-accuracy instruction following.

Built for agentic systemsBuilt for agentic systems

Nemotron 3 Super combines a hybrid Transformer–Mamba architecture with mixture-of-experts routing to improve compute efficiency while maintaining strong reasoning performance.

Key characteristics:

• 120B parameters, 12B active;
• Up to 1M token context;
• Multi-Token Prediction for faster long-form generation;
• Open weights, open datasets and open training recipes;
• Text-in, text-out model inference.

The model targets production use cases such as:

• Software development workflows, including code generation and analysis;
• Deep research agents for long-horizon planning and reasoning;
• Financial document processing;
• Cybersecurity triage and threat intelligent analysis.

Run NVIDIA Nemotron 3 Super in productionRun NVIDIA Nemotron 3 Super in production

On Nebius Token Factory, Nemotron 3 Super can be deployed via:

• Dedicated GPU endpoints with guaranteed performance;
• Autoscaling throughput for production workloads;
• OpenAI-compatible API integration;
• EU or US regional deployment options;
• Optional zero-retention inference.

Token Factory enables teams to move from model access to production deployment without managing GPU clusters or inference infrastructure.

Get startedGet started

Nemotron 3 Super is available today in the Nebius Token Factory console.

Deploy via API or test in the Playground. Build AI engineered for your product.