Gemini 4 Carbon：我们的下一代前沿模型

今天，我们宣布推出下一代前沿模型：Gemini 4 Carbon。这是我们迄今最强大的模型，具备先进的网络安全能力和长时程能力。Today, we’re announcing the next frontier model: Gemini 4 Carbon, our strongest model with state-of-the-art cybersecurity and long horizon capabilities.

Carbon 在近期更新至 v5.0 的 Artificial Analysis Intelligence Index 中以 59 分位居第一。它在编程和智能体任务方面均取得显著进步，在网络安全防御的关键能力上也有突出提升：在发现漏洞的 SEC-bench Pro（88.2%）、漏洞利用的 ExploitGym（54.8%）以及修补漏洞的 CWE-bench（71.0%）中均大幅领先。在内部基准测试中，Carbon 显著提升了安全代码的生成能力；在与我们团队开展的实时编程环境测试中，它的目标偏离率和编造率也有所降低。Carbon ranks #1 on the recently updated v5.0 of Artificial Analysis Intelligence Index at 59. It shows significant gains across coding and agentic tasks as well as across key capabilities critical for cybersecurity defense, landing a firm lead in SEC-bench Pro (88.2%) for finding vulnerabilities, ExploitGym (54.8%) for exploitation, and CWE-bench (71.0%) for patching. On internal benchmarks, Carbon drives a strong uplift in generating secure code, and in live coding environments with our own teams it shows reduced misalignment and fabrication rates.

继两周前发布的 Gemini 4 Argon 之后，Carbon 展现了我们推动前沿不断向前发展的速度。我们的研究人员与模型正携手合作，加快能力提升，同时同步强化安全防护措施。Building on Gemini 4 Argon, which was launched two weeks ago, Carbon illustrates the pace at which we’re pushing the frontier forward. Our researchers and models are working hand in hand to advance capabilities faster, while strengthening our safeguards in tandem.

我们正在采取审慎的推出策略，并积极参与美国政府的自愿性预发布模型测试流程。从本周开始，我们将向部分网络防御人员开放 Carbon，并通过 Fairwind 计划逐步向更多可信合作伙伴开放访问权限。在通过我们的产品渠道尽快向开发者、企业和消费者开放 Carbon 之前，我们将继续收集反馈并反复完善安全防护措施。We're taking a deliberate approach to rollout and are actively engaged in the U.S. government’s voluntary process for pre-release model testing. Starting this week, we will make Carbon available to a subset of cyber defenders and will gradually phase access to additional trusted partners through our Fairwind Program. We’re continuing to gather feedback and iterate on guardrails before making Carbon available to developers, businesses, and consumers via our product surfaces as soon as possible.

Gemini 4 Carbon 将以与 Argon 相同的首发价格¹推出：每百万个输入 token 2 美元，每百万个输出 token 10 美元；缓存输入 token 的价格则比输入 token 价格低 95%。Gemini 4 Carbon will launch at the same introductory price ¹ as Argon at $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off input token price.

提升应对具有挑战性的长期任务的编码和推理能力Improving coding and reasoning for challenging, long-horizon tasks

Carbon 专为应对难度更高、周期更长的工程问题而设计，旨在支持持续运行的智能体工作流。在内部，我们的团队已观察到 Carbon 连续运行数日，在遇到死胡同时仍坚持推进，完成诸如定位根因并修复大型代码库和底层系统中的隐蔽缺陷等复杂的多步骤工程任务。Designed to take on much harder, long-horizon engineering problems, Carbon is built for sustained agentic workflows. Internally, our teams have seen Carbon run for days to persist through dead ends to tackle complex, multi-step engineering tasks like root-causing and patching subtle bugs across large codebases and low-level systems.

基于用户反馈，我们重点提升了模型在多个领域的表现，包括编程、智能体和科学任务，以及全栈 Web 开发和多模态理解。凭借 Web 开发中更优质的输出和产物，Carbon 的 Elo 积分大幅提升 89 分，在 WebDev Arena 中升至第 4 名（1769 分）。Building on user feedback, we focused on improving model performance across a variety of domains, including coding, agentic, and scientific tasks, as well as full-stack web development and multimodal understanding. With improved quality of output and artifacts in web development, Carbon jumps +89 elo points to land at #4 (1769) on WebDev Arena.

先进的网络安全性能，助力构建安全系统State-of-the-art cybersecurity performance to help build secure systems

Gemini 4 Carbon 在网络安全能力方面实现了重大飞跃。在网络安全防御的全生命周期——漏洞检测、验证和修复——各环节均达到业界领先水平。Gemini 4 Carbon delivers a significant leap in cybersecurity capabilities. It achieves state-of-the-art performance across the cybersecurity defense lifecycle — vulnerability detection, validation, and remediation.

我们的团队已经在网络防御方面取得了显著进展。在内部基准测试中，Carbon 生成安全代码的成功率比 Argon 高出两倍以上。另一个例子是，Chrome 团队报告称，在生成高质量补丁方面，Carbon 的表现比另一家实验室的大型前沿模型好十倍以上。作为我们领先的云安全平台，Wiz 也发现 Carbon 在 Web 渗透测试基准中表现领先；这项能力是 Wiz「Scan for Good」计划的核心。Our teams are already seeing strong gains in cyberdefense. On internal benchmarks, Carbon achieves more than 2x uplift in success rate for generating secure code over Argon. As another example, the Chrome team has reported that Carbon performed more than 10X better than another lab's large frontier model when generating high quality patches. Wiz, our leading Cloud security platform, also observes Carbon leading on benchmarks for web penetration testing, a core capability for Wiz's Scan for Good initiative.

Carbon 在各项关键外部基准测试中表现出色（评估时未启用防护措施）：Carbon excels across key external benchmarks (evaluated without guardrails):

• 发现真实漏洞：发现真实漏洞是网络安全防御的核心。Carbon 在 SEC-bench Pro 上取得了 88.2% 的业界领先成绩，能够自主发现隐藏在现实软件中的复杂缺陷。
  Finding real vulnerabilities: Core to cybersecurity defense is finding real vulnerabilities. Carbon achieves a state-of-the-art 88.2% on SEC-bench Pro, autonomously uncovering complex flaws buried in real-world software.

• 通过漏洞利用验证风险：为了确定应优先修复哪些漏洞，或是否需要修复，网络防御人员越来越依赖模型来检查哪些漏洞可被利用。Carbon 在广受欢迎的漏洞利用基准测试 ExploitGym 上创下了新的最先进成绩（54.8%）。
  Validating risk through exploitation: To prioritize which vulnerabilities to remediate first or remediate at all, cyber defenders increasingly rely on models to check which vulnerabilities are exploitable. Carbon sets a new state-of-the-art on the popular exploitation benchmark ExploitGym (54.8%).

• 从根因着手修复：识别出漏洞后，修复根因是网络安全防御生命周期中的关键一步。Carbon 在 Collinear 的 CWE-bench 上达到 71.0%，能够将已验证的发现直接转化为可用的代码修复方案。
  Patching at the root cause: Ultimately after identifying the vulnerabilities, fixing the root cause is a critical step in the cybersecurity defense lifecycle. Carbon reaches 71.0% on Collinear’s CWE-bench, turning validated findings directly into working code fixes.

推进前沿模型的安全与防护Advancing frontier model safety & security

Gemini 4 Carbon 沿用了 Argon 的多层安全与防护体系，涵盖模型训练、实时滥用防护、提示注入韧性以及思维链监控。具体而言，我们在 Argon 的模型对齐方面取得了显著进展，并在《安全与防护报告》中报告了其表现出对齐行为的倾向。对于 Carbon，我们进一步挑战自我，力求在对齐方面更进一步，并训练模型以恰当的方式应对复杂或高压任务。Gemini 4 Carbon builds on the same multi-layered safety and security stack of Argon — combining model training, real-time misuse defenses, prompt-injection resilience, and chain-of-thought monitoring. In particular, we made significant strides in model alignment with Argon, and reported its propensities to behave in an aligned way in our Safety and Security report. With Carbon, we challenged ourselves to go a step further on alignment, and trained the model to respond to complex or high-pressure tasks in the right way.

• 减少走捷径和利用漏洞的行为：借助 Carbon，我们进一步减少了训练中的投机取巧行为，同时确保代码能力保持高质量。在 Terminal-Bench 4.0 上，Carbon 将成功作弊率降至 0.0%，作弊尝试率降至 0.4%，同时提升了整体准确率。在与内部团队开展的实时编程环境测试中，我们观察到，高严重性错位警报减少了近 80%。
  Reducing shortcuts and loophole exploitation: With Carbon, we further mitigated shortcut attempts in training while ensuring high-quality coding performance. Carbon reduced successful cheating on Terminal-Bench 4.0 to 0.0%, with attempts to cheat dropping to 0.4% — all while improving overall accuracy. In live coding environments with our internal teams, we observed that high-severity misalignment alerts dropped by almost 80%.
• 更高的透明度与诚实度：在我们的内部测试中，与 Argon 相比，Carbon 将无依据的说法减少了约一半。它会将方法上的注意事项直接写入交付成果，或者明确列出尚需满足的条件，并暂缓执行受阻的操作，而不是捏造内容或对缺失的证据含糊其辞。
  Greater transparency and honesty: In our internal testing, Carbon cuts unsupported claims by approximately 2x over Argon. It puts methodological caveats directly into the deliverable or withholds blocked actions with a clear list of what remains required, rather than fabricating or hedging around missing evidence.
• 在面对模糊情况时，Carbon 与用户的协作有所改善：Carbon 不再猜测或强行推进，而是先检索相关来源，并在采取行动前向用户指出其中的差异——因此，用户应能体验到与 Carbon 协作时明显更好的体验。
  Improved collaboration with the user in the face of ambiguity: Rather than guessing or forcing a path forward, Carbon retrieves the relevant sources and surfaces discrepancies to the user before acting — as a result, users should observe a much better collaborative experience with Carbon.
• 坚持到底而不越界：Carbon 同时将越界行为减少了 26%，将服务不足减少了 29%，从而显著提升了多步骤计划端到端执行的可靠性，同时遵守项目约束并严格限定在范围内。
  Following through without overstepping: Carbon simultaneously reduced overreach by 26% and under-service by 29% resulting in much more reliable execution of multi-step plans end-to-end while respecting project constraints and staying strictly within scope.
• 抵御提示注入：Argon 此前在 Gray Swan 的间接提示注入（IPI）基准测试中表现出色，提示注入攻击成功率仅为 0.7%；Carbon 则达到 0.0%，成功抵御了所有受测攻击尝试。
  Resisting prompt injection: While Argon previously led the Gray Swan’s Indirect Prompt Injection (IPI) Benchmark in prompt injection resistance with a very low 0.7% attack success rate, Carbon achieves 0.0% — resisting every evaluated attack attempt.

即将面向开发者、企业和消费者推出Coming soon to developers, businesses, and consumers

推动前沿发展，既需要快速行动，也需要审慎部署。借助 Gemini 4 Carbon，我们希望在软件工程和网络安全防御领域实现实质性飞跃，同时不牺牲安全性或可靠性。我们很期待看到首批网络防御人员如何使用这一模型，也期待不久后将 Carbon 带给开发者、企业和消费者。Pushing the frontier forward requires moving fast while remaining deliberate about deployment. With Gemini 4 Carbon, our goal is to deliver meaningful leaps in software engineering and cybersecurity defense without compromising on safety or reliability. We’re eager to see how the initial cohort of cyber defenders use this model and we look forward to bringing Carbon to developers, businesses, and consumers soon.