NVIDIA Releases Nemotron 3 Super: A 120B Parameter Open-Source Hybrid Mamba-Attention MoE Model Delivering 5x Higher Throughput for Agentic AI

2026-03-11 · Source: Machine Learning ML & Generative AI News · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Advanced, quick

Summary

NVIDIA has released Nemotron 3 Super, a 120-billion parameter open-source model designed for multi-agent reasoning and enterprise-grade autonomous agents. This model features a hybrid Mixture-of-Experts (MoE) architecture, integrating both Mamba and Transformer layers, and boasts a 1-million token context window. It achieves 7x higher throughput and double the accuracy compared to its predecessor, making it efficient for complex, long-form tasks. Nemotron 3 Super also introduces "Reasoning Budgets," enabling developers to manage compute costs by adjusting between deep-search analysis and low-latency responses. NVIDIA has open-sourced the entire training stack, including weights and datasets, to promote transparency and advanced AI development.

Key takeaway

For AI Architects and CTOs evaluating models for agentic AI, Nemotron 3 Super offers a compelling open-source option with its 120B parameters and hybrid MoE design. Its "Reasoning Budgets" feature allows precise control over compute costs, which is critical for deploying efficient enterprise-grade autonomous agents. Consider integrating this model for applications requiring high throughput and accuracy in complex, long-form tasks.

Key insights

Nemotron 3 Super is an open-source 120B parameter hybrid MoE model for efficient multi-agent reasoning.

Principles

• Hybrid architectures enhance performance
• Open-sourcing fosters transparency
• Context windows improve long-form task efficiency

Method

Nemotron 3 Super combines Mamba and Transformer layers in a MoE architecture, utilizing "Reasoning Budgets" for cost-controlled inference.

In practice

• Use for enterprise autonomous agents
• Control compute costs with "Reasoning Budgets"
• Leverage 1-million token context for long tasks

Topics

• Nemotron 3 Super
• Mixture-of-Experts
• Mamba-Attention Hybrid
• Agentic AI
• Open-Source Models

Best for: AI Architect, CTO, VP of Engineering/Data, AI Engineer, Machine Learning Engineer, Software Engineer

Related on AIssential

• Introducing Nemotron 3 Super: An Open Hybrid Mamba-Transformer MoE for Agentic Reasoning — Nemotron 3 Super optimizes multi-agent AI with a hybrid MoE architecture, 1M-token context, and NVFP4 pretraining.
• NVIDIA’s New AI Just Changed Everything — NVIDIA's Nemotron 3 Super offers a powerful, open-source AI assistant with significant speed improvements via novel architectural and mathematical techniques.
• Nemotron 3: NVIDIA’s Latest LLM in Plain English — Nemotron 3 balances advanced AI capabilities with practical deployment efficiency for agentic workloads.
• New NVIDIA Nemotron 3 Super Delivers 5x Higher Throughput for Agentic AI — Nemotron 3 Super is an open 120B-parameter model designed for scalable, accurate agentic AI systems.
• NVIDIA AI Releases Nemotron 3 Ultra: An Open 550B Mixture-of-Experts Hybrid Mamba-Transformer for Long-Running Agents — Nemotron 3 Ultra is an open hybrid Mamba-Transformer MoE designed for efficient, long-context agentic AI with flat decode costs.
• NVIDIA Releases Nemotron-Cascade 2: An Open 30B MoE with 3B Active Parameters, Delivering Better Reasoning and Strong Agentic Capabilities — NVIDIA's Nemotron-Cascade 2 MoE model excels in reasoning and agentic tasks via Cascade RL and MOPD.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning ML & Generative AI News.