{
  "slug": "deepseek-v3-2",
  "order": 27,
  "name": "DeepSeek V3.2 (671B)",
  "baseTitle": "DeepSeek V3.2",
  "company": "DeepSeek",
  "country": "China",
  "flag": "🇨🇳",
  "date": "2025-12-01",
  "decoderType": "Sparse MoE",
  "attention": "MLA with DeepSeek Sparse Attention",
  "layerMix": "61 MLA",
  "scale": "671B total, 37B active (5.5% active)",
  "totalParams": 671000000000,
  "activeParams": 37000000000,
  "context": "128,000",
  "contextTokens": 128000,
  "license": "MIT License",
  "kvCacheText": "68.6 KiB",
  "kvCacheKiB": 68.6,
  "kvCacheLevel": "Low",
  "summary": "DeepSeek's successor keeps the V3 template but adds sparse attention to cut long-context costs.",
  "keyDetail": "Adds sparse attention while retaining the DeepSeek V3-style MTP path.",
  "image": {
    "full": "https://sebastianraschka.com/llm-architecture-gallery/images/architectures/deepseek-v3-2-671b.webp",
    "thumb": "https://sebastianraschka.com/llm-architecture-gallery/images/architectures/thumbnails/deepseek-v3-2-671b.webp",
    "alt": "Figure 48: DeepSeek V3.2",
    "width": 960,
    "height": 939
  },
  "articleUrl": "https://magazine.sebastianraschka.com/p/the-big-llm-architecture-comparison#%C2%A716-deepseek-v32",
  "links": [
    {
      "label": "View in article",
      "href": "https://magazine.sebastianraschka.com/p/the-big-llm-architecture-comparison#%C2%A716-deepseek-v32",
      "title": null
    },
    {
      "label": "config.json",
      "href": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/blob/main/config.json",
      "title": "deepseek-ai/DeepSeek-V3.2"
    },
    {
      "label": "License",
      "href": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2/blob/main/LICENSE",
      "title": null
    },
    {
      "label": "Tech report",
      "href": "https://arxiv.org/pdf/2512.02556",
      "title": null
    }
  ],
  "concepts": [
    {
      "slug": "mla",
      "label": "MLA",
      "name": "Multi-Head Latent Attention",
      "url": "https://llm-atlas-7or.pages.dev/concepts/mla/"
    },
    {
      "slug": "moe",
      "label": "MoE",
      "name": "Mixture of Experts",
      "url": "https://llm-atlas-7or.pages.dev/concepts/moe/"
    },
    {
      "slug": "deepseek-sparse-attention",
      "label": "DSA",
      "name": "DeepSeek Sparse Attention",
      "url": "https://llm-atlas-7or.pages.dev/concepts/deepseek-sparse-attention/"
    },
    {
      "slug": "mtp",
      "label": "MTP",
      "name": "Multi-Token Prediction",
      "url": "https://llm-atlas-7or.pages.dev/concepts/mtp/"
    }
  ],
  "aai": {
    "v4": {
      "label": "AA Intelligence Index v4.3.2",
      "total": 16,
      "profile": {
        "Scientific": 6
      },
      "url": "https://artificialanalysis.ai/models/deepseek-v3-2"
    }
  },
  "sourceUrl": "https://sebastianraschka.com/llm-architecture-gallery/#card-deepseek-v3-2",
  "family": "moe",
  "url": "https://llm-atlas-7or.pages.dev/models/deepseek-v3-2/",
  "markdownUrl": "https://llm-atlas-7or.pages.dev/models/deepseek-v3-2.md"
}