{
  "slug": "nanbeige-4-2-3b",
  "order": 91,
  "name": "Nanbeige 4.2 (3B)",
  "baseTitle": "Nanbeige 4.2 3B",
  "company": "BOSS Zhipin",
  "country": "China",
  "flag": "🇨🇳",
  "date": "2026-07-21",
  "decoderType": "Dense",
  "attention": "GQA with RoPE",
  "layerMix": "22 shared GQA layers × 2 passes",
  "scale": "4B total, 3B non-embedding",
  "totalParams": 4000000000,
  "activeParams": 4000000000,
  "context": "262,144",
  "contextTokens": 262144,
  "license": "Apache License 2.0",
  "kvCacheText": "176 KiB",
  "kvCacheKiB": 176,
  "kvCacheLevel": "High",
  "summary": "Compact agentic model that runs the same 22-layer dense transformer stack twice, resulting in 44 transformer blocks (layers) without duplicating the layer weights.",
  "keyDetail": "Reuses the same 22-layer stack for a second pass, giving 44 block applications with 3B non-embedding parameters.",
  "image": {
    "full": "https://sebastianraschka.com/llm-architecture-gallery/images/architectures/nanbeige-4-2-3b.webp",
    "thumb": "https://sebastianraschka.com/llm-architecture-gallery/images/architectures/thumbnails/nanbeige-4-2-3b.webp",
    "alt": "Figure 1: Nanbeige 4.2 3B",
    "width": 960,
    "height": 953
  },
  "articleUrl": null,
  "links": [
    {
      "label": "config.json",
      "href": "https://huggingface.co/Nanbeige/Nanbeige4.2-3B/blob/main/config.json",
      "title": "Nanbeige/Nanbeige4.2-3B"
    },
    {
      "label": "Tech report",
      "href": "https://huggingface.co/Nanbeige/Nanbeige4.2-3B/blob/main/Nanbeige42_report.pdf",
      "title": null
    }
  ],
  "concepts": [
    {
      "slug": "looped-depth-sharing",
      "label": "Looped Transformer",
      "name": "Looped Transformer",
      "url": "https://llm-atlas-7or.pages.dev/concepts/looped-depth-sharing/"
    },
    {
      "slug": "gqa",
      "label": "GQA",
      "name": "Grouped-Query Attention",
      "url": "https://llm-atlas-7or.pages.dev/concepts/gqa/"
    }
  ],
  "aai": null,
  "sourceUrl": "https://sebastianraschka.com/llm-architecture-gallery/#card-nanbeige-4-2-3b",
  "family": "dense",
  "url": "https://llm-atlas-7or.pages.dev/models/nanbeige-4-2-3b/",
  "markdownUrl": "https://llm-atlas-7or.pages.dev/models/nanbeige-4-2-3b.md"
}