谷歌发布前沿模型 Gemini 4 Carbon：网络安全迎来“重大飞跃”，登顶多项基准评测

Google 发布了其最新前沿模型 Gemini 4 Carbon，在编程、Agent 任务和网络安全方面带来了“显著提升”。在周三的一项公告中，Google 重点强调了 Carbon 在 Artificial Analysis Intelligence Index (v5.0) 基准测试中位居榜首，领先于 Claude Opus 5.5 和 GPT-6 Astra。

新的 Gemini 4 Carbon 模型在 Google 本月早些时候发布的 Argon 基础上实现了进一步升级。该模型目前仅面向选定的一批网络安全防御公司开放，但未来将通过其 Fairwind 计划逐步扩展至更多合作伙伴。

Google 表示，Carbon 体现了“我们推进前沿技术的步伐”，在网络防御领域带来了“重大飞跃”，使其在挖掘漏洞和生成安全代码方面表现更佳。Carbon 目前在衡量 AI 模型发现现实世界漏洞能力的基准测试 SEC-bench Pro，以及评估模型如何将漏洞转化为 Exploit 攻击代码的 ExploitGym 上均位列第一。

据 Google 介绍，Gemini 4 Carbon 专为完成更长周期的任务而打造，并补充称其内部团队曾目睹该模型“能够连续运行数天，在碰壁走入死胡同后依然坚持探索，以解决复杂的多步骤工程任务”。此外，Google 还表示 Carbon 更擅长避免走捷径、抵御 Prompt 注入攻击，并能更好地与用户开展协作。与 Argon 相比，Carbon 将“无依据的主张”减少了一半，从而能够提供更高的“透明度与诚实度”。

该模型的首发定价与 Argon 同样具有竞争力，费用为每百万输入 Token 2 美元，每百万输出 Token 10 美元。

Google has released its latest frontier model, Gemini 4 Carbon, offering “significant gains” in coding, agentic tasks, and cybersecurity. In an announcement on Wednesday, Google highlights Carbon’s top spot on the Artificial Analysis Intelligence Index (v5.0) benchmark, with Claude Opus 5.5 and GPT-6 Astra trailing behind.

The new Gemini 4 Carbon model offers improvements over Argon, which Google released earlier this month. It’s currently only available to a select group of cybersecurity defense companies, but it will eventually expand to other partners in its Fairwind Program.

The company says Carbon reflects the “pace at which we’re pushing the frontier forward,” coming with a “significant leap” in cyberdefense that makes it better at finding vulnerabilities and generating secure code. Carbon currently holds the number one spot on SEC-bench Pro, a benchmark that measures an AI model’s ability to discover real-world bugs, and ExploitGym, which measures how a model can turn a vulnerability into exploits.

Google says Gemini 4 Carbon is built for completing longer tasks, adding that its internal teams have seen it “run for days to persist through dead ends to tackle complex, multi-step engineering tasks.” It’s also better at avoiding shortcuts, resisting prompt injection attacks, and collaborating with users, according to Google. Carbon is supposed to offer more “transparency and honesty” as well, by cutting “unsupported claims” by two times when compared to Argon.

It will launch at the same competitive introductory price as Argon, at $2 per million input tokens and $10 per million output tokens.