{
  "slug": "nemotron-3-ultra-550b-a55b",
  "order": 78,
  "name": "Nemotron 3 Ultra (550B-A55B)",
  "baseTitle": "Nemotron 3 Ultra 550B-A55B",
  "company": "NVIDIA",
  "country": "United States",
  "flag": "🇺🇸",
  "date": "2026-06-04",
  "decoderType": "Hybrid MoE",
  "attention": "Mostly Mamba-2 with a few GQA layers",
  "layerMix": "12 GQA + 48 Mamba-2 + 48 MoE",
  "scale": "550B total, 55B active (10% active)",
  "totalParams": 550000000000,
  "activeParams": 55000000000,
  "context": "1,000,000",
  "contextTokens": 1000000,
  "license": "NVIDIA Nemotron Open Model License",
  "kvCacheText": "12 KiB",
  "kvCacheKiB": 12,
  "kvCacheLevel": "Very low",
  "summary": "NVIDIA's Ultra model scales the Nemotron 3 hybrid stack to 550B total parameters while keeping 55B active per token.",
  "keyDetail": "Scales the latent-MoE hybrid to 108 layers with 48 Mamba-2, 48 latent MoE, and 12 GQA layers.",
  "image": {
    "full": "https://sebastianraschka.com/llm-architecture-gallery/images/architectures/nemotron-3-ultra-550b-a55b.webp",
    "thumb": "https://sebastianraschka.com/llm-architecture-gallery/images/architectures/thumbnails/nemotron-3-ultra-550b-a55b.webp",
    "alt": "Figure 1: Nemotron 3 Ultra 550B-A55B",
    "width": 960,
    "height": 931
  },
  "articleUrl": null,
  "links": [
    {
      "label": "config.json",
      "href": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4/blob/main/config.json",
      "title": "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4"
    },
    {
      "label": "License",
      "href": "https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-nemotron-open-model-license/",
      "title": null
    },
    {
      "label": "Tech report",
      "href": "https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Ultra-Technical-Report.pdf",
      "title": null
    }
  ],
  "concepts": [
    {
      "slug": "gqa",
      "label": "GQA",
      "name": "Grouped-Query Attention",
      "url": "https://llm-atlas-7or.pages.dev/concepts/gqa/"
    },
    {
      "slug": "moe",
      "label": "MoE",
      "name": "Mixture of Experts",
      "url": "https://llm-atlas-7or.pages.dev/concepts/moe/"
    },
    {
      "slug": "hybrid-attention",
      "label": "Hybrid Attention",
      "name": "Hybrid Attention",
      "url": "https://llm-atlas-7or.pages.dev/concepts/hybrid-attention/"
    }
  ],
  "aai": {
    "v5": {
      "label": "AA Intelligence Index v5.0",
      "total": 11.8,
      "profile": {
        "Knowledge Work": 12.9,
        "Coding": 0.9,
        "Science": 6,
        "General": 27.4
      },
      "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b"
    },
    "v4": {
      "label": "AA Intelligence Index v4.3.2",
      "total": 22.9,
      "profile": {
        "General": 34.1,
        "Scientific": 15.8,
        "Coding": 20.4,
        "Agents": 18.5
      },
      "url": "https://artificialanalysis.ai/models/nvidia-nemotron-3-ultra-550b-a55b"
    }
  },
  "sourceUrl": "https://sebastianraschka.com/llm-architecture-gallery/#card-nemotron-3-ultra-550b-a55b",
  "family": "hybrid-moe",
  "url": "https://llm-atlas-7or.pages.dev/models/nemotron-3-ultra-550b-a55b/",
  "markdownUrl": "https://llm-atlas-7or.pages.dev/models/nemotron-3-ultra-550b-a55b.md"
}