---
{
  "availableLangs": [
    "zh",
    "ar",
    "de",
    "en",
    "es",
    "fr",
    "hi",
    "id",
    "it",
    "ja",
    "ko",
    "ms",
    "nl",
    "pt",
    "ru",
    "th",
    "tr",
    "vi",
    "zh-hant"
  ],
  "categories": [
    "category-science-technology"
  ],
  "createdAt": "2026-10-30T00:00:00+08:00",
  "creators": [
    "Google DeepMind"
  ],
  "id": "work_gemini_4_carbon",
  "lang": "zh",
  "published": {
    "day": 28,
    "display": "2026年10月28日",
    "month": 10,
    "precision": "day",
    "year": 2026
  },
  "revision": 19,
  "slug": "gemini-4-carbon",
  "sources": [
    {
      "author": "Google",
      "date": "2026-10-28",
      "id": "s1",
      "title": "Gemini 4 Carbon: our next frontier model",
      "url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/"
    },
    {
      "author": "9to5Google",
      "date": "2026-10-28",
      "id": "s2",
      "title": "Google announces Gemini 4 Carbon with ‘significant gains across coding and agentic tasks’",
      "url": "https://9to5google.com/2026/10/28/gemini-4-carbon-announcement/"
    },
    {
      "author": "Ars Technica",
      "date": "2026-10-28",
      "id": "s3",
      "title": "Google announces Gemini 4 Carbon, its latest frontier AI model that you can’t use",
      "url": "https://arstechnica.com/ai/2026/10/google-announces-gemini-4-carbon-its-latest-frontier-ai-model-that-you-cant-use"
    }
  ],
  "status": "published",
  "summary": "Google DeepMind 于 2026 年 10 月 28 日发布的前沿人工智能模型，侧重网络安全防御与长周期任务，初期仅向受信任的网络安全人员开放。",
  "tags": [],
  "title": "Gemini 4 Carbon",
  "type": "work",
  "updatedAt": "2026-10-30T14:58:20+08:00",
  "updatedBy": "user:admin",
  "workKind": "ai_model"
}
---

# Gemini 4 Carbon

Google DeepMind 于 2026 年 10 月 28 日发布的前沿人工智能模型，侧重网络安全防御与长周期任务，初期仅向受信任的网络安全人员开放。

Gemini 4 Carbon 是 [Google DeepMind](/zh/organizations/google-deepmind.html) 于 2026 年 10 月 28 日发布的前沿人工智能模型，[Google](/zh/organizations/google.html) 称其为迄今能力最强的模型，重点提升网络安全防御与长周期任务处理能力。

## 发布与开放范围

Carbon 与 9 月 30 日发布的 [Gemini 4 Argon](/zh/works/gemini-4-argon.html) 沿用同一套安全与防护措施。发布时仅向一小批经筛选的网络安全防御人员开放，随后通过 Fairwind 计划扩大至更多受信任合作伙伴，再面向开发者、企业和消费者推出。Google 表示正在参与美国政府的前沿模型发布前自愿测试流程。

## 评测结果

Artificial Analysis Intelligence Index v5.0 给出 59 分，Carbon 在该指数中排名第一；WebDev Arena 得分 1768，排名第四。Google 公布的网络安全评测在未启用防护栏的条件下进行：SEC-bench Pro 漏洞发现得分 88.2%，ExploitGym 漏洞可利用性验证得分 54.8%，CWE-bench（Collinear）根因修复得分 71.0%。Google 同时称，生成安全代码的成功率较 Argon 提升一倍以上，高严重度失配警报减少近 80%，Gray Swan 间接提示注入攻击成功率为 0.0%。

## 定价

介绍期价格与 Argon 相同，为每百万输入 token 2 美元、输出 token 10 美元，缓存输入按输入价的 95% 折扣计费；介绍期结束后为每百万输入 token 4 美元、输出 token 20 美元。

更新时间：2026-10-30

## 标签


## 关系

- 出版方：[Google DeepMind](/zh/organizations/google-deepmind.html) 

第 1 / 1 页，共 1 条。

## 来源

- Gemini 4 Carbon: our next frontier model：https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/ 
- Google announces Gemini 4 Carbon with ‘significant gains across coding and agentic tasks’：https://9to5google.com/2026/10/28/gemini-4-carbon-announcement/ 
- Google announces Gemini 4 Carbon, its latest frontier AI model that you can’t use：https://arstechnica.com/ai/2026/10/google-announces-gemini-4-carbon-its-latest-frontier-ai-model-that-you-cant-use