NVIDIA Launches Nemotron 3 Super — 5x Faster Agentic AI Model
120B parameter open model with hybrid Mamba-Transformer architecture delivers 5x higher throughput and 2x accuracy over predecessor. Only 12B parameters active at inference via MoE. 1M token context window for multi-agent workflows. Powers Perplexity search, CodeRabbit, Greptile integrations. Available on Hugging Face, Perplexity, OpenRouter. - Why it matters: Major efficiency breakthrough for agentic AI systems. Open weights with full methodology released. - Links: https://blogs.nvidia.com/blog/nemotron-3-super-agentic-ai/ | https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-FP8