Google DeepMind Launches EmbeddingGemma 2 Multimodal Embedding Model

Google DeepMind Launches EmbeddingGemma 2 Multimodal Embedding Model Google DeepMind has introduced EmbeddingGemma 2, an open-source, 740-million-parameter multimodal embedding model built on the Gemma 4 architecture under an Apache 2.0 license. The model maps text, code, audio, images, and video into a shared embedding space designed for on-device inference. EmbeddingGemma 2 features 740 million total parameters in a modular design, requiring 270M for text-only tasks alongside optional vision (170M) and audio (300M) encoders. The model incorporates an 8K token context window capable of handling up to 5.5 minutes of audio, 29 images, or 58 video frames. It uses Matryoshka Representation Learning to compress 768-dimension vectors down to 128 dimensions, enabling up to a 6x storage reduction. Quantized deployments on a Google Pixel 11 Pro require approximately 191MB active RAM for text-only weights and around 567MB for the full multimodal model. Architecture and Efficiency EmbeddingGemma 2 uses technology derived from Gemini Embedding models and Gemma 4. Its modular structure allows developers to run text-only setups at 270 million parameters or attach vision and audio components up to 740 million parameters. Through Matryoshka Representation Learning, output dimensions can be reduced from 768 to 512, 256, or 128, cutting storage and memory demands by up to sixfold. Memory optimizations allow the quantized full model to operate within 567MB of active RAM on devices like the Pixel 11 Pro. Performance and Cross-Modal Capabilities The model features an 8K token context window—four times larger than the first EmbeddingGemma—enabling cross-modal searches like querying audio using text or locating video segments via voice memos. On benchmark evaluations, it scored 78.68 on MTEB Code, up from 68.76, while recording top scores on the Massive Audio Embedding Benchmark (MAEB) for sub-1B parameter models. On-Device Deployment and Ecosystem Designed for offline, privacy-first retrieval augmented generation (RAG) and semantic search, EmbeddingGemma 2 shares a text tokenizer and audio encoder with Gemma 4 to minimize memory when run concurrently. The weights are distributed on Hugging Face and Kaggle under an Apache 2.0 license, supporting deployment tools such as LiteRT, MediaPipe, MLX, llama.cpp, Ollama, vLLM, and WebGPU.

Google DeepMind 发布开源轻量多模态嵌入模型 EmbeddingGemma 2 Google DeepMind 正式推出基于 Gemma 4 架构的 EmbeddingGemma 2。该模型参数量为 7.4 亿，采用商业友好的 Apache 2.0 许可证开源，支持将文本、代码、图像、音频与视频统一映射至同一嵌入空间，专门针对端侧推理进行了高效优化。 总参数量 7.4 亿，采用模块化设计：仅文本工作负载只需 2.7 亿参数，可选配 1.7 亿参数视觉编码器与 3 亿参数音频编码器。 采用俄罗斯套娃表示学习（MRL），输出维度可从 768 维截断至 512、256 或 128 维，最高可减少 6 倍存储与内存占用。 具备 8K token 上下文窗口，支持处理最长 5.5 分钟音频、29 张图像或 58 帧视频及其交错组合。 经量化后在 Google Pixel 11 Pro 上运行，纯文本权重仅需约 191MB 活跃 RAM，完整多模态模型约需 567MB。 跨模态统一与模块化端侧优化 EmbeddingGemma 2 构建于 Gemma 4 架构并基于 Gemini Embedding 技术，支持原生多模态融合，可实现通过语音备忘录检索视频片段或根据文本搜索音频录音。模型具备极高的模块化灵活性，全模态版本包含 7.4 亿参数，但纯文本场景可精简至 2.7 亿参数运行。借助 MRL 技术，向量存储可动态降维压缩，量化后在移动硬件上占用极低活跃内存。 性能突破与长上下文处理能力 该模型在保持多语言文本能力的同时，MTEB Code 得分提升 9.92 点至 78.68，在代码检索、图像、视频和音频基准测试（如 MAEB）中树立了 10 亿参数以下模型的新标准，部分表现超越了两倍于其体量的专用模型。此外，其上下文长度从前代的 2K 扩展至 8K tokens，支持单次处理长达 5.5 分钟的音频或数十帧视觉内容。 端侧 RAG 与广泛生态集成 由于与 Gemma 4 共享文本分词器及音频编码器，两者可联合部署组成端侧 RAG 检索生成管线并节省总体内存占用。模型现已在 Hugging Face 与 Kaggle 提供权重下载，并深度支持 MediaPipe、LiteRT、WebGPU、vLLM、llama.cpp、Ollama、Qdrant 及 Unsloth 等多种端侧推理与微调开发工具。

Oct 06, 2026 | EmbeddingGemma 2 is the most capable model for on-device multimodal embeddings, natively mapping combinations of text, images, audio, and video into a unified embedding space. Sahil Dua Research Engineer, Google DeepMind Henrique Schechter Vera Research Engineer, Google DeepMind Your browser does not support the audio element. Listen to article [[duration]] minutes This content is generated by Google AI. Generative AI is experimental We introduced EmbeddingGemma last year to provide a lightweight option for high-quality text embeddings, to help your apps organize, search, and connect information directly on consumer hardware. The developer community’s response blew past our expectations. With more than 20 million downloads, builders have used it to power smarter on-device search tools and privacy-first retrieval augmented generation (RAG) pipelines. Today, we’re launching EmbeddingGemma 2 , expanding beyond text to unify code, images, video, and audio in a shared embedding space. Built on the Gemma 4 architecture and released under a commercially permissive Apache 2.0 license, EmbeddingGemma 2 has 740 million parameters, making it optimal for on-device inference. It can help find a specific video clip from a voice memo, or search through hours of audio recordings based on a text query, all processed by a single, natively multimodal model. Built from the same technology as Gemini Embedding models, EmbeddingGemma 2 is: Best-in-class for its size: Achieves leading scores among sub-1B multimodal embedders for its size across benchmarks like MTEB (Massive Text Embedding Benchmark) Code and MAEB (Massive Audio Embedding Benchmark), while matching or outperforming many larger models across text, vision, and audio tasks. Modular by design: Requires as little as 270M parameters for text-only workloads with optional vision (170M) and audio (300M) encoders for full multimodal support. Storage-efficient: Using Matryoshka Representation Learning (MRL), developers can dynamically truncate output vectors from 768 dimensions down to 512, 256, or 128 dimensions. This provides up to 6x storage reduction for local vector databases and memory usage. Optimized for on-device performance: Runs efficiently within tight resource constraints. With quantization, on a Google Pixel 11 Pro, EmbeddingGemma 2 requires as little as ~191MB active RAM for text-only weights and ~567MB for the full multimodal model. Extended context ready: Features an 8K token context window (4x larger than EmbeddingGemma 1), allowing it to process up to 5.5 minutes of audio, 29 images, 58 video frames, or interleaved combinations thereof directly on local hardware. Achieving top-tier quality for code, vision, and audio EmbeddingGemma 2 matches the strong multilingual text performance of EmbeddingGemma while delivering a significant 9.92-point improvement on code performance (in MTEB Code, from 68.76 to 78.68), making it well-suited for local codebase indexing, semantic code search, and coding agent retrieval. Across image, video, documents, and audio, it sets a new standard in quality-per-parameter for sub-1B models and even outperforms some specialist models more than twice its size. Find full evaluation metrics and model information in the EmbeddingGemma 2 model card . Enabling semantic search, routing, and retrieval, fully on-device EmbeddingGemma 2 brings robust capabilities directly to edge hardware. Generating embeddings locally helps ensure data privacy, reduces pipeline latency, and empowers developers to build cross-modal search and retrieval that works entirely offline. When paired with generative models such as Gemma 4, EmbeddingGemma 2 enables on-device RAG pipelines that understand complex multimodal data. Because EmbeddingGemma 2 is built on Gemma 4 and shares its text tokenizer and audio encoder, developers can run both models together in a unified pipeline with a lower combined total memory footprint. To learn how to build on-device search and RAG systems with LiteRT, read the Google AI Edge blog post . Getting started with EmbeddingGemma 2 We worked closely with the following partners to ensure EmbeddingGemma 2 works immediately where you build: Download the models: Find the model weights on Hugging Face and Kaggle , with Gemini Enterprise Agent Platform Model Garden availability coming soon. Visit LiteRT Community on Hugging Face for models optimized for on-device. On-device deployment: Develop cross-platform apps with Google AI Edge MediaPipe for turnkey embedding, retrieval & decision tasks or LiteRT for custom model integration. Build for the browser with transformers.js or WebGPU . Use your favorite development tools : Serve the model efficiently using transformers, sentence-transformers, MLX , vLLM, llama.cpp , SGLang, Ollama , and LMStudio. Store your embedding vectors with Qdrant . Fine-tuning: Follow guidance by Unsloth for how to fine-tune EmbeddingGemma 2 for your use cases. Explore our developer guide , documentation , and guides for inference and fine-tuning .