{
  "$type": "site.standard.document",
  "bskyPostRef": {
    "cid": "bafyreifo4cgmu6qbhhugu7xrqsz5xhbycjkfjso5j676a65dxbk522giwi",
    "uri": "at://did:plc:mysqw426hguigpyif4r2ndsx/app.bsky.feed.post/3milmehol56f2"
  },
  "coverImage": {
    "$type": "blob",
    "ref": {
      "$link": "bafkreihn7wla25elqpxbvso2g36she6kql3rjz3qeu3twg7gkpk7iyjazy"
    },
    "mimeType": "image/webp",
    "size": 94878
  },
  "path": "/google-gemma-4-open-source-ai-model-benchmark-specs-analysis/",
  "publishedAt": "2026-04-03T06:53:46.000Z",
  "site": "https://www.5news.com.hk",
  "tags": [
    "科技",
    "全部"
  ],
  "textContent": "Google 最新推出嘅 Gemma 4 開源 AI 模型系列，官方基準測試（Benchmark）成績正式曝光。數據顯示，Gemma 4 喺邏輯推理、編程及多模態處理上，效能大幅拋離上一代 Gemma 3 模型。新系列提供 E2B、E4B、26B MoE 及 31B 四款尺寸，全面涵蓋邊緣運算到高階工作站。本文將詳細為大家整合各型號嘅硬件規格與跑分表現，拆解呢款新世代模型嘅實際應用潛力。",
  "title": "Google Gemma 4 效能實測報告：跑分大幅拋離上代，31B 與 26B MoE 規格詳解"
}