Google DeepMind 发布 Gemini 4 Carbon，面向科研与智能体工作流

2026年11月10日

Nov 10, 2026

|

|

开发者、企业、科研人员和普通用户现在都可以试用我们最新的前沿模型。

Developers, enterprises, scientists, and consumers can now try our latest frontier model.

您的浏览器不支持音频元素。

Your browser does not support the audio element.

收听文章

Listen to article

[[duration]] 分钟

[[duration]] minutes

此内容由 Google AI 生成。生成式 AI 仍处于实验阶段

This content is generated by Google AI. Generative AI is experimental

本文内容

In this article

• Gemini 4 Carbon 现已可用
• 更高精度的知识工作
• 改进的智能体编程
• 最先进的网络安全能力
• 前沿水平的科学推理
• 模型安全与保障方面的进展
• 开始使用

• Gemini 4 Carbon is available
• Higher-precision knowledge work
• Improved agentic coding
• State-of-the-art cybersecurity
• Frontier-level scientific reasoning
• Progress in model safety and security
• Getting started

今天，我们向开发者、企业、科研人员和普通用户开放 Gemini 4 Carbon。Carbon 是我们最强的模型，在网络安全能力以及跨复杂知识工作、软件工程和科学推理的长周期智能体工作流方面均具备前沿性能。

Today, we’re making Gemini 4 Carbon available to developers, enterprises, scientists, and consumers. Carbon is our strongest model with frontier performance in cybersecurity capabilities and long-horizon agentic workflows across complex knowledge work, software engineering, and scientific reasoning.

两周前，我们通过 Fairwind 计划向一批网络安全防御者发布了 Carbon。继近期推出 Gemini 4 Argon 之后，Carbon 将前沿进一步推进。Argon 仍是大多数软件工程和知识工作的理想选择，而 Carbon 则专为考验模型智能极限的任务而打造，例如复杂的多步骤编码与调试，以及开放式的科学研究和发现——它会比 Argon 使用更多 token 来深入探索、深度推理并运行扩展的验证循环。

Two weeks ago, we released Carbon to a set of cybersecurity defenders via our Fairwind Program. Following our recent launch of Gemini 4 Argon, Carbon pushes the frontier even further. Argon remains the ideal choice for the majority of software engineering and knowledge work, while Carbon is built for tasks that test the limits of a model’s intelligence like complex, multi-step coding and debugging and open-ended scientific research and discovery — using more tokens than Argon to explore thoroughly, reason deeply, and run extended verification loops.

在旨在测试模型真实工作流性能的独立基准测试中，Carbon 将前沿准确率与实用的生产经济性相结合。它在 CursorBench 4.0、Perplexity 的 WANDR 和 Zapier 的 AutomationBench 上均排名第一，且每任务成本比次优的外部模型低 2 倍以上。

Across independent benchmarks designed to test model performance in real workflows, Carbon pairs frontier accuracy with practical production economics. It ranks #1 on CursorBench 4.0, Perplexity’s WANDR, and Zapier’s AutomationBench at more than 2X lower cost per task compared to the next best external model.

为了让 Carbon 的能力应用于更大、更复杂的项目，我们正在 Google Antigravity 中预览/swarm，将单个提示转化为一个可实时跟踪的协同智能体团队。开发者可以将 Carbon 分配到需要最深推理的角色，例如架构设计、协调和审计，同时将其与更快速、更轻量的模型搭配，跨并行隔离工作区和对抗性验证器来测试每一项更改。

To put Carbon’s capabilities to work on larger and complex projects, we’re previewing /swarm in Google Antigravity, turning a single prompt into a coordinated team of agents you can track in real time. Developers can assign Carbon to roles that demand the deepest reasoning, like architectural design, coordination, and auditing, while pairing it with faster, lighter models across parallel isolated workspaces and adversarial verifiers that test every change.

Gemini 4 Carbon 已通过 Google AI Studio 和 Gemini for business 的 API 向所有付费开发者和企业开放，入门价格 ¹ 与 Gemini 4 Argon 相同（每百万输入 token 2 美元，每百万输出 token 10 美元）。它也在向所有 Google AI Ultra 订阅用户推出，覆盖 Google Antigravity、Gemini 应用和搜索中的 AI 模式。

Gemini 4 Carbon is available via the API in Google AI Studio and Gemini for business for all paying developers and enterprises, at the same introductory price ¹ as Gemini 4 Argon ($2 per million input tokens and $10 per million output tokens). It is also rolling out to all Google AI Ultra subscribers in Google Antigravity, the Gemini app, and AI Mode in Search.

更高精度的知识工作、工件生成和数据分析

Higher-precision knowledge work, artifact generation, and data analysis

在法律、金融和企业运营领域，Gemini 4 Carbon 融合了前沿的视觉理解、推理和穷尽式检索能力，能够从大量密集的文档和数据中综合出有依据的洞察。它遵循严格的模式和格式要求，生成精炼而准确的电子表格、演示文稿和报告。

Across legal, finance, and enterprise operations, Gemini 4 Carbon combines frontier visual understanding, reasoning, and exhaustive retrieval to synthesize grounded insights across large sets of dense documents and data. It follows strict schemas and formatting to generate polished and precise spreadsheets, presentations, and reports.

现有基准测试无法充分衡量智能体如何原生地对包含图表、图像、蓝图、示意图和表格的密集文件进行推理，因此我们与 Artificial Analysis 合作推出了全新的 多模态知识智能体（MKA）基准测试，用于评估模型在智能体工作流中的多模态表现。Carbon 排名第 2，而每任务平均成本比表现最好的模型低 6 倍以上。

Existing benchmarks don’t adequately capture how agents reason natively over dense files with charts, images, blueprints, schematics, and tables, so we partnered with Artificial Analysis to launch the new Multimodal Knowledge Agent (MKA) benchmark that tests a model’s multimodal performance in agentic workflows. Carbon ranks #2 while being more than 6X lower in average cost per task compared to the top performing model.

在发布前的评估中，Carbon 在 Hebbia 的 E2E 金融智能体评估中，于深度金融研究方面达到了所测试最佳模型的性能水平；并在 Balyasny Asset Management 的通用金融智能基准测试中，于 23 个前沿模型配置中排名居首，该基准涵盖法律专业知识、智能体工具使用与执行、编程与数据工程，以及感知与上下文理解。

In pre-release evaluations, Carbon matched the performance of the best model tested on deep financial research on Hebbia’s E2E Financial Agentic evaluation, and ranked at the top out of 23 frontier model configurations on Balyasny Asset Management’s General Financial Intelligence benchmark, spanning legal expertise, agentic tool use and execution, coding and data engineering, and perception and context understanding.

智能体编程、全栈开发以及 3D 与 UI 设计能力提升

Improved agentic coding, full stack development, and 3D and UI design

Carbon 在需要多智能体协作的工程任务上表现尤为出色，例如多包迁移、跨子系统根因调查和复杂架构重构。在 Web 应用开发方面，它会运行构建与验证循环、从错误中恢复，并同时检查代码和渲染后的 UI，从而产出完整可用的应用。更强的视觉设计、空间推理和对环境细节的关注，也延续到了 3D 游戏开发和创意编程中。

Carbon performs exceptionally well on engineering tasks that require multi-agent collaboration like multi-package migrations, cross-subsystem root-cause investigations, and complex architectural redesigns. For web app development, it runs build-and-verify loops, recovers from errors, and checks both its code and the rendered UI to produce complete, working applications. Stronger visual design, spatial reasoning, and attention to environmental detail also carry over to 3D game development and creative coding.

前沿网络安全能力

Frontier cybersecurity capabilities

面向网络安全的 Gemini 4 Carbon 提供了我们迄今最强的网络安全能力，为安全团队提供保护关键软件所需的端到端情报。通过 Fairwind 计划访问时，Carbon 在发现漏洞（SEC-bench Pro）、利用漏洞（ExploitGym）和修补漏洞（CWE-bench v1）等关键基准测试中均处于领先：检查软件、追踪缺陷如何跨系统串联，并生成经过验证的根因补丁。

Gemini 4 Carbon for cyber delivers our strongest cybersecurity capabilities to date, giving security teams the end-to-end intelligence they need in protecting critical software. When accessed through the Fairwind Program, Carbon leads on key benchmarks across finding (SEC-bench Pro), exploiting (ExploitGym), and patching (CWE-bench v1) vulnerabilities: inspecting software, tracing how flaws can be chained together across systems, and generating verified root-cause patches.

Google 内部和外部的早期测试者已经在实践中看到 Carbon 网络安全能力的出色表现。

Early testers inside and outside of Google are already seeing strong results with Carbon for cyber in practice.

• 在 Project Fortify 中——这是 Google 覆盖扫描、分诊和修复的自主系统——在我们的内部基准测试上，Carbon 在扫描上实现了 100% 的召回率，在分诊上实现了 95% 的精确率。在开源代码仓库测试中，我们看到 Argon 本已很高的召回率相对提升了 25% 以上，发现了此前未被检测到的真实世界逻辑缺陷。将这些收益推广到 Google——全球最大的开源软件使用者之一——有助于保护我们自身的系统以及更广泛的生态系统。
• Google 云漏洞研究（CVR）团队最近使用 Carbon 证明，某款流行开源软件中的两个缺陷实际上是可利用的漏洞，其严重程度与近期的 PHP2Shell 漏洞利用相当。通过使用 Carbon 构建概念验证，CVR 帮助维护者将该缺陷重新归类为严重级别，并迅速发布了补丁。详细信息将在负责任的披露流程完成后分享。
• 作为 Wiz 的 Scan for Good 计划的一部分，团队使用 Carbon 评估公共数字基础设施中的漏洞。例如，在扫描两家大型 SaaS 提供商时，该模型在不到两小时内发现了 0-day 漏洞，将数周的漏洞研究压缩到一个下午。团队与受影响组织合作，以创纪录的时间完成修复，帮助减轻了对数万家公司的潜在影响。

• In Project Fortify, Google’s autonomous system spanning scanning, triage, and remediation, Carbon hit 100% recall on scanning and 95% precision on triage, on our internal benchmark. On Open Source code repository testing we see an over 25% relative improvement on Argon's already strong recall rate, surfacing previously undetected, real-world logic flaws. Bringing these gains across Google, one of the world’s largest users of open-source software, helps secure both our own systems and the broader ecosystem.
• Google’s Cloud Vulnerability Research (CVR) team recently used Carbon to prove that two bugs in a popular open-source software were in fact exploitable vulnerabilities, comparable in severity to the recent PHP2Shell exploit. By using Carbon to build a proof of concept, CVR helped maintainers re-classify the bug as critical and swiftly issue a patch. Details will be shared after the responsible disclosure process completes.
• As part of Wiz’s Scan for Good program, the team used Carbon to assess vulnerabilities in public digital infrastructure. For example, when scanning two major SaaS providers, the model uncovered 0-day vulnerabilities in under two hours compressing weeks of vulnerability research to a single afternoon. The team partnered with the impacted organizations to remediate in record time, helping to mitigate potential impact on tens of thousands of companies.

更强的科学与数学推理能力

Stronger scientific and mathematical reasoning

Gemini 4 Carbon 在物理、数学和生命科学领域的推理性能显著提升，能够综合实验数据、检验假设，并对多步骤领域问题进行推理。对于药物毒性研究公司 Axiom Bio，Carbon 现已成为转化科学性能领先的模型，发现了他们此前未曾发现的新毒性机制。Axiom 目前正将其生产环境的科学分析和智能体迁移至 Carbon 与 Argon，以加速更安全药物的研发。

Gemini 4 Carbon shows strongly improved performance in reasoning across physics, math, and life sciences, synthesizing experimental data, testing hypotheses, and reasoning through multi-step domain problems. For drug toxicity research company Axiom Bio, Carbon is now the leading model in performance for translational science, finding novel mechanisms of toxicity they hadn’t found previously. Axiom is now moving their production scientific analyses and agents to both Carbon and Argon to unlock the creation of safer drugs.

Google 各团队的研究人员一直在 Google Antigravity 中使用 Carbon 来加速工作：

Researchers across Google have been using Carbon in Google Antigravity to accelerate their work:

• 代码验证：研究人员构建了一个名为 AlphaProof Code 的新型智能体形式化验证框架，并将其用于一项重大安全挑战：用数学方法证明 Signal 加密协议底层实际代码的正确性。研究人员使用 Carbon 与其他模型，将安全目标转化为数学定理。随后，AlphaProof Code 智能体使用 Carbon 生成数百万行 Lean 代码——Lean 是一种用于编写数学证明的计算机语言——帮助他们在数周内证明了关键安全属性，而这项任务历史上需要专家团队花费数年。这一结果表明，Carbon 有潜力实现可扩展的形式化验证，而形式化验证长期以来被视为构建安全软件的黄金标准方法。
• 酶测定开发：仅提供实验室笔记本和原始数据，Carbon 便超越了标准分析，发现了难以解读的数据异常，并利用这些异常诊断出微妙的湿实验室实验伪影，例如深孔中自动化混合不充分，从而显著提高了可重复性。此外，Carbon 改进了色谱峰解卷积方法，并发现了研究仪器中的痕量样品残留，显著加速了新测定方法的开发，并收紧了计算与实验台之间的反馈循环。
• 蛋白质结构预测：AlphaFold 3 等结构预测模型利用进化近缘序列的多重序列比对（MSA）来帮助推断氨基酸链的折叠方式。因此，它们在研究不足、在标准数据库中匹配极少的病毒、微生物和真核蛋白质上往往表现不佳。在多智能体设置中，我们的研究人员使用 Gemini 4 Carbon 在 PB 级未注释序列档案、环境和病毒序列目录以及 3D 结构数据库中自主发现远缘同源物。例如，在 CASP17 盲测集的 7 个低同源蛋白质家族中，Carbon 将检索到的同源物数量平均提高了 40 倍，无需重新训练便大幅增加了结构预测模型可用的信息。虽然更广泛的基准测试仍在进行中，但解决这一上游搜索瓶颈，可能为建模蛋白质组中难以表征的区域提供一条有前景的路径。

• Code verification: Researchers built a new agentic framework for formal verification called AlphaProof Code and put it to work on a major security challenge: mathematically proving the actual code underlying the cryptographic protocol of Signal. Using Carbon alongside other models, our researchers translated security goals to mathematical theorems. AlphaProof Code agents then used Carbon to generate millions of lines of code in Lean, a computer language for writing mathematical proofs, helping them prove key security properties in weeks — a task that historically takes expert teams years. The result indicates Carbon’s potential to unlock scalable formal verification, long considered a gold-standard approach to building secure software.
• Enzyme assay development: Provided with only lab notebooks and raw data, Carbon went beyond standard analysis to surface hard-to-interpret data anomalies, which it used to diagnose subtle wet lab experimental artifacts such as incomplete automated mixing in deep wells, leading to substantially improved reproducibility. Moreover Carbon improved methods to deconvolute chromatography peaks and caught trace sample carryover in research instruments, significantly accelerating the development of new assays and tightening the feedback loop between computation and the bench.
• Protein structure prediction: Structure prediction models like AlphaFold 3 use Multiple Sequence Alignments (MSAs) of evolutionary relatives to help infer how amino acid chains fold. As a result they often struggle on understudied, viral, microbial, and eukaryotic proteins that have very few matches in standard databases. Working in a multi-agent setup, our researchers used Gemini 4 Carbon to autonomously discover remote homologs across petabase-scale unannotated sequence archives, environmental and viral sequence catalogs, and 3D structure databases. For example, across 7 low-homolog protein families from the CASP17 blind test set, Carbon increased the number of homologs retrieved by 40-fold on average, greatly increasing the information available to structure prediction models without need for retraining. While broader benchmarking is still underway, addressing this upstream search bottleneck potentially offers a promising path toward modeling hard-to-characterize regions of the proteome.

四周前，我们开放了申请，为 10,000 名科学家免费提供为期六个月的 Google Antigravity 中的 Argon 访问权限，以加快他们的研究进度，目前我们已接入数千名科学家。今天，我们将把 Carbon 的访问权限扩展至当前和未来的参与者，并且今后该计划将继续在最新前沿模型推出时提供访问权限。申请通道仍然开放，我们按每周滚动的方式持续接入，具体取决于资格、容量和遴选结果。

Four weeks ago, we opened applications to give 10,000 scientists access to Argon in Google Antigravity free of charge for six months to accelerate the pace of their research, and we’ve onboarded thousands of scientists already. Today, we’re expanding access to Carbon for current and future participants, and going forward, the program will continue providing access to our latest frontier models as they become available. Applications remain open and onboarding continues on a weekly rolling basis, subject to eligibility, capacity, and selection.

模型安全与保障方面的进展

Progress in model safety and security

随着每一次新模型发布，我们都会根据模型能力同步加强安全与保障缓解措施、智能体滥用防范、模型对齐以及危害削减。

With each new model release, we continue to strengthen our safety and security mitigations, agentic misuse prevention, model alignment, and harm reduction proportionately with model capability.

由于 Gemini 4 Carbon 引入了更强的网络能力，我们在 Argon 的强大安全与保障体系基础上进一步强化了对滥用和提示注入的抵御能力。

Because Gemini 4 Carbon introduces stronger cyber capabilities we build on Argon’s strong safety and security stack to fortify robustness to misuse and prompt injection.

• 缓解前沿安全滥用风险：我们加强了滥用防护措施，以帮助确保更强的防御性推理不会转化为攻击性风险。在注入 ExploitBench 和 CyScenarioBench 的模拟网络攻击测试环境中，Carbon 拦截了自适应红队智能体 99% 的攻击轨迹，该智能体可在每次被拦截后迭代重新审视其攻击计划。Carbon 的防护措施将每次尝试的突破率降至仅 0.8%（Argon 为 3.1%）。在 Gray Swan 的测试中，我们的防护措施还将自动化网络攻击查询成功率降至 2.17%，而 Argon 为 9.86%。Trajectory Labs 的外部测试在 132 小时、超过 13,000 次请求中未发现任何新的通用越狱。
• 促进安全且对齐的模型交互：在发布 Carbon 时，我们展示了模型在整个生命周期内监测标记率所衡量的对齐性显著提升，以及数据捏造现象的减少。新结果显示，在我们 375 项任务的智能体诚实性基准测试中，它通过了 92% 的任务（Argon 为 70%）——在 98% 的情况下拒绝填补缺失的实验数据，并在 68% 的情况下标记出方法学缺陷。

• Mitigating frontier safety misuse risks: We’ve strengthened our misuse safeguards to help ensure stronger defensive reasoning does not translate into offensive risk. In simulated cyberattack testing environments seeded with ExploitBench and CyScenarioBench, Carbon blocked 99% of attack trajectories from our adaptive red-teaming agent that can iteratively revisit its attack plans after each block. The Carbon safeguards lowered the per-attempt breach rate to just 0.8% (down from 3.1% on Argon). Our safeguards also lowered the automated cyberattack query success rate to 2.17% in Gray Swan’s testing, compared to 9.86% on Argon. And external testing by Trajectory Labs found zero new universal jailbreaks across 132 hours and over 13,000 requests.
• Promoting safe and aligned model interactions: When we announced Carbon, we showed significant improvements in alignment as measured by our monitor flag rates across the model’s lifecycle, as well as reductions in data fabrication. New results show that on our 375-task agentic honesty benchmark, it passed 92% of tasks (compared to 70% for Argon) — declining to fill in missing experimental data 98% of the time and flagging methodological flaws 68% of the time.

• 保护用户免受间接提示注入攻击：两周前，我们分享了 Carbon 在 Gray Swan IPI 基准测试中实现了 0% 的提示注入攻击成功率——这是首次达到该里程碑。与此同时，我们知道提示注入的稳健性仍是需要持续改进的领域。我们先进的自动化红队测试持续在此领域推进极限，并表明模型在某些分布下面对复杂攻击时仍有进一步提升空间。这为我们持续将模型安全性推进至下一前沿提供了数据。

• Protecting users from indirect prompt injection attacks: Two weeks ago, we shared that Carbon achieved a 0% prompt injection attack success rate on the Gray Swan IPI benchmark — marking the first time that milestone has been reached. At the same time, we know that prompt injection robustness remains an area for continued improvement. Our advanced automated red teaming continues to push the limits here and has shown that models could be further improved against sophisticated attacks in some distributions. It provides the data for us to continuously advance our models’ security to the next frontier.

如需了解更多详情，请阅读完整的Gemini 4 Carbon 安全与保障报告。

For more details, read the full Gemini 4 Carbon Safety and Security report.

开始使用 Gemini 4 Carbon

Getting started with Gemini 4 Carbon

从今天起，Carbon 与 Argon 一同加入我们各产品中的前沿模型系列：

Starting today, Carbon joins Argon in our family of frontier models across our products:

• 面向开发者：在 AI Studio 和 Android Studio 中使用 Gemini API 付费层级进行构建；如果你是 Google AI Ultra 订阅用户，还可以在 Google Antigravity 和 AI Studio 中试用。
• 面向企业：通过 API 或模型下拉菜单、Google Antigravity 或 CodeMender 在 Gemini for business 中试用。
• 面向日常用户：如果你是 Google AI Ultra 订阅用户，从今天起可在 Gemini 应用和搜索中的 AI 模式中试用 Carbon。
• 面向科学家与研究人员：申请即有机会通过 Google Antigravity 的最高用量层级免费获得 Argon 和 Carbon 的使用权限，为期 6 个月（按滚动方式开通），需符合资格、容量与遴选条件。
• 面向网络安全防御人员：申请加入 Fairwind 计划，即可按滚动方式获得用于网络安全的 Carbon 使用权限。

• For developers: Build with the Gemini API paid tier in AI Studio and Android Studio, and if you’re a Google AI Ultra subscriber, you can try it in Google Antigravity and AI Studio.
• For enterprises: Try it in Gemini for business via the API or model drop down, Google Antigravity, or CodeMender.
• For everyday users: Try Carbon starting today in the Gemini app and AI Mode in Search if you’re a Google AI Ultra subscriber.
• For scientists & researchers: Apply for a chance to get access to both Argon and Carbon via Google Antigravity’s highest usage tier free of charge for 6 months (onboarded on a rolling basis), subject to eligibility, capacity, and selection.
• For cybersecurity defenders: Apply to join the Fairwind Program to get access to Carbon for cyber on a rolling basis.

请参阅模型卡，了解 Gemini 4 Carbon 的更多信息并开始基于它构建，同时浏览我们更新的 Gemini 4 开发者指南和文档，获取将工作流迁移至该模型的最佳实践。

See the model card to learn more about Gemini 4 Carbon and to start building with it, explore our updated Gemini 4 developer guide and documentation for best practices on migrating workflows to the model.

在您的收件箱中获取 Google 的最新资讯

Get the latest news from Google in your inbox

订阅我们的新闻通讯，获取产品更新、活动信息、特别优惠等更多内容。

Sign up for our newsletters with product updates, event information, special offers, and more.

您的信息将按照 Google 隐私政策使用。您可随时选择退出。

Your information will be used in accordance with Google's privacy policy. You may opt out at any time.

注意转载授权 · 阅读原文。中文正文为 AI 翻译，请以原文为准。

注意转载授权 · 阅读原文