Gemini Embedding 2 is now generally available.
Summary (EN)
Google announced the general availability of Gemini Embedding 2 through the Gemini API and Vertex AI, moving its natively multimodal embedding model from preview into production use. The company says the model is designed to let developers and enterprises search, compare, and reason across text, image, video, and audio within a single embedding system, instead of stitching together separate pipelines for each modality. In the post, Google points to prototypes built during preview, including advanced e commerce discovery engines and more efficient video analysis workflows, as evidence that organizations want multimodal retrieval and matching infrastructure that can support real applications rather than just experimentation. By taking the model to general availability, Google is signaling that it has reached the stability and optimization level required for production deployment, an important threshold for teams deciding whether to build on a platform or continue with custom infrastructure. The release also fits a broader trend in which embedding systems are becoming core application primitives for recommendation, search, ranking, retrieval augmented generation, and media understanding. For commercial teams, the practical takeaway is that multimodal search infrastructure is becoming easier to buy as a managed service, which lowers the cost and complexity of deploying AI features in retail, media, enterprise knowledge systems, and other settings where users interact with mixed data types. Google presents Gemini Embedding 2 not as a research demo, but as a production layer for multimodal application development.
Summary (ZH)
Google 宣布 Gemini Embedding 2 正式进入 GA，可通过 Gemini API 与 Vertex AI 使用，意味着这款原生多模态嵌入模型从预览阶段转入面向生产环境的正式可用状态。Google 表示，该模型的目标是在单一嵌入系统中同时处理文本、图像、视频与音频的搜索、匹配和推理需求，帮助开发者和企业避免为不同模态分别搭建割裂的数据处理与检索管线。官方在文章中提到，预览阶段已有团队利用它构建先进的电商发现引擎以及更高效的视频分析流程，这被用来证明市场对多模态检索与理解基础设施的真实需求已经超出原型实验层面。此次转为 GA，代表 Google 认为模型的稳定性和性能优化已足以支撑正式业务部署，这对正在评估是否采用托管式 AI 平台的企业来说是一个重要信号。从产业趋势看，嵌入模型正在成为推荐、搜索、排序、RAG 检索和媒体理解等应用的底层基础组件。对商业应用团队而言，更直接的意义在于，多模态搜索能力正越来越容易以平台化服务形式采购，从而降低零售、媒体、企业知识系统等场景上线 AI 功能的复杂度与成本。Google 在这里强调的不是一项研究展示，而是一个可直接投入实际产品开发的多模态应用基础层。
Source