Paper

Gemini Embedding 2: A Native Multimodal Embedding Model from Gemini

Rothe, Sascha, Chen, Yichang, Barnes, Megan, Heigold, Georg, Sung, Yunhsuan, Karukas, Stephen, Agrawal, Ayush, Yang, Jizheng, Goenka, Sonam, Vyas, Nidhi, Wang, Emma, Pchelin, Vladimir, Yi-Ting, Chen, Jiao, Junyi, Hegde, Chaitra, Hengxuan Ying, Liu, Chaoren, Hoffmann, Raphael, Ding, Zhongli, Mosley, Keegan, Boratko, Michael, Gleicher, Zach, Dua, Sahil, Montes, Alberto, Choi, Min, Yin, Qin, Zhang, Grace, Halley Fede, Lee, Jinhyuk, Shih-Cheng, Huang, Poulet, Kevin, Nath, Dev, Zhang, Shijie, Potetz, Brian, Hess, Andreas, Russo, Sebastian, Meng, Rui, Loher, Lucia, Hui, Hui, Twu, Alice, Cer, Daniel, Reveillon, Antoine, Schlattner, Philippe, Duerig, Tom, Han, Shuoxuan, Salz, Daniel, Allauzen, Cyril, Gu, Yang, Suganathan, Paul, Qiu, Steve, Kim, Dahun, Samari, Babak, Yang, Zhen, Bagby, Tom, Chen, Kaifeng, Santana, Roberto, Gill, Karan, Frank Palma Gomez, Kulkarni, Sujay, Ma, Ji, Zhou, Wenlei, Gao, Shen, Seyedhosseini, Mojtaba, Zheng, Jack, Kumar, Shankar, Zhang, Hesen, Hauth, Anja, Wu, Jiaxing, Yang, Albert, Li, Zhe, Gustavo Hernández Ábrego, Dabral, Tanmaya, Bhai, Siddharth, Andonov, Jovan, Zhang, Wangyuan, Zhang, Shanfeng, Lowe, Wing, Rao, Vik

2026

Overview

We introduce Gemini Embedding 2, a native multimodal embedding model that allows embedding video, audio, image, and text modalities in a unified representation space. We leverage the multimodal capabilities of Gemini to produce embeddings for arbitrary combinations of interleaved inputs across all these modalities that generalize well across a wide variety of tasks. Applying large-scale contrastive learning in a multi-task multi-stage training setup, we achieve state-of-the-art performance on key embedding benchmarks including unimodal, cross-modal, and multimodal retrieval spanning a diverse set of tasks. We show that our embedding model demonstrates strong performance (with a score of 62.9 R@1 on MSCOCO, 68.8 NDCG@10 on Vatex, 69.9 on MTEB multilingual and 84.0 on MTEB Code) across a variety of tasks surpassing the performance of specialized models. These unified capabilities make Gemini Embedding 2 a promising candidate for downstream use cases such as RAG, recommendation and search. Furthermore, its robust zero-shot performance across distinct fields - from astronomy and bioscience to fine arts and the culinary arts - establishes it as a highly reliable, out-of-the-box representation even for specialized domains.

• Publisher: Cornell University Library, arXiv.org
• Subject: Astronomy / Contrastive learning / Embedding / Representations