{
  "contentStorage": "postgres",
  "title": "Gemini 1.0 Nano",
  "description": "Google DeepMind 于 2023 年 12 月 6 日 发布的 Gemini 1.0 Nano。",
  "url": "/ja/works/gemini-1-0-nano.html",
  "entity": {
    "id": "work_gemini_1_0_nano",
    "type": "work",
    "title": "Gemini 1.0 Nano",
    "slug": "gemini-1-0-nano",
    "summary": "Google DeepMind 于 2023 年 12 月 6 日 发布的 Gemini 1.0 Nano。",
    "status": "published",
    "categories": [
      "category-science-technology"
    ],
    "tags": [],
    "sources": [
      {
        "id": "s1",
        "title": "Introducing Gemini: Google's most capable AI model yet",
        "url": "https://blog.google/technology/ai/google-gemini-ai/"
      },
      {
        "id": "s2",
        "title": "Google Pixel December feature drop: New Gemini AI-powered updates",
        "url": "https://blog.google/products/pixel/pixel-feature-drop-december-2023/"
      },
      {
        "id": "s3",
        "title": "Gemini (language model) - Wikipedia",
        "url": "https://en.wikipedia.org/wiki/Gemini_(language_model)"
      },
      {
        "id": "s4",
        "title": "Gemini: A Family of Highly Capable Multimodal Models",
        "url": "https://arxiv.org/abs/2312.11805"
      }
    ],
    "createdAt": "2026-11-11T13:43:47+08:00",
    "updatedAt": "2026-11-11T13:43:47+08:00",
    "updatedBy": "user:admin",
    "revision": 1,
    "body": "## 发布背景\n\nGemini 1.0 Nano 是 [[organization_google_deepmind|Google DeepMind]] 于 2023 年 12 月 6 日 发布的 Gemini 模型[^s1]。该模型分 Nano-1（18 亿参数）与 Nano-2（32.5 亿参数）两个版本，均经 4-bit 量化，首发于 Pixel 8 Pro，用于录音摘要与 Gboard 智能回复。[^s1]\n\n## 定价与获取\n\n官方未公布Gemini 1.0 Nano的公开按 token 计价[^s1]。官方未公布其公开 API 标识[^s1]。\n\n## 性能表现\n\n以下均为 Google DeepMind 在发布公告与技术报告中公布的厂商自报成绩，尚未经独立复现：\n\n- MMLU (5-shot)：55.8%，厂商自报，Nano-2 得分；同表 Nano-1 为 45.9%[^s1]。\n- MATH (4-shot)：22.8%，厂商自报，Nano-2 得分；同表 Nano-1 为 13.5%[^s1]。\n- BoolQ：79.3%，厂商自报，Nano-2 得分；同表 Nano-1 为 71.6%[^s1]。\n- MBPP：27.2%，厂商自报，Nano-2 得分；同表 Nano-1 为 20.0%[^s1]。\n\n## 主要特点\n\n上下文窗口 32,768 token[^s1]。\n\n该模型仅通过设备端运行，无公开 API 与定价，数据无需离开手机即可离线使用[^s1]。\n\n最大输出 token 数未见官方公布；上下文 32,768 token 为技术报告的训练序列长度，设备端实际可用窗口可能更小[^s1]。\n\n## 定位\n\nGemini 1.0 家族中最小的端侧档位，低于同代 Pro 与 Ultra，其端侧定位由后续世代的 Gemini Nano 迭代继承。该模型已退役。",
    "url": "/works/gemini-1-0-nano.html",
    "lang": "ja",
    "sourceLang": "zh",
    "fallback": true,
    "availableLangs": [
      "zh"
    ],
    "workKind": "ai_model"
  },
  "entities": null,
  "relations": [
    {
      "id": "relation_work_gemini_1_0_nano_published_by_organization_google_deepmind",
      "fromId": "work_gemini_1_0_nano",
      "relationType": "published_by",
      "toId": "organization_google_deepmind",
      "sourceRefs": [
        {
          "entityId": "work_gemini_1_0_nano",
          "sourceId": "s1"
        }
      ],
      "label": "Published by",
      "targetId": "organization_google_deepmind",
      "targetTitle": "Google DeepMind",
      "targetUrl": "/ja/organizations/google-deepmind.html",
      "incoming": false
    }
  ],
  "versions": [
    {
      "revision": 1,
      "updatedAt": "2026-11-11T13:43:47+08:00",
      "message": "当前版本",
      "hasSnapshot": true,
      "url": "/ja/works/gemini-1-0-nano/at/1.html"
    }
  ],
  "lang": "ja",
  "availableLangs": [
    "zh"
  ],
  "fallback": true,
  "infoFields": [
    {
      "key": "work_kind",
      "label": "Work kind",
      "value": "AIモデル"
    }
  ],
  "relationsTotal": 1,
  "relationCounts": [
    {
      "relationType": "published_by",
      "incoming": false,
      "label": "Published by",
      "total": 1
    }
  ]
}