Google DeepMind 推出 Carbon 模型，面向长时间推理任务
Google 这次把 Carbon 同时推给 Ultra 订阅用户、API 开发者和企业三条线，还能挖软件漏洞，安全团队和需要跑长推理任务的开发者可以重点看看 AI Studio 的接入方式。
Google DeepMind 宣布推出新模型 Carbon，主打需要持续推理的任务，能力覆盖多模态密集文件综合、构建交互式 3D 网页体验以及挖掘软件漏洞三类场景。
分发渠道分为三层：Gemini App 和 Search 的 AI Mode 面向 Google AI Ultra 订阅用户开放；Gemini 4 Carbon 通过 API 登陆 Google AI Studio 和 Gemini for business，所有付费开发者和企业均可调用；Gemini for business、CodeMender 和 Google Antigravity 则面向企业用户。
从产品布局看，Google 将 Carbon 同时塞进消费端（Ultra 订阅）、开发者端（AI Studio API）和企业端（Antigravity、CodeMender），其中漏洞挖掘与 CodeMender 组合指向安全场景，与 Anthropic、OpenAI 在长程推理模型上的竞争进一步加剧。
From synthesizing dense multimodal files to building interactive 3D web experiences, or even uncovering software vulnerabilities, Carbon helps solve problems that require sustained reasoning. Now rolling out in: 🔘 The @GeminiApp and AI Mode in Search for Google AI Ultra subscribers. 🔘 Gemini 4 Carbon is available via the API in @GoogleAIStudio and Gemini for business for all paying developers and enterprises 🔘 Gemini for business, CodeMender, and Google Antigravity for enterprise us goo.gle/48Jd6ud G0w
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Carbon, Gemini, Google DeepMind, 推理模型.
Source attribution: curated from Google DeepMind (@GoogleDeepMind). Cross-source corroboration: 11 independent sources have covered this development (including Anthropic, 宝玉, The Rundown AI), confirming industry-wide signal strength.
