Gemini Embedding 2 is now generally available.

Source: Google Date: 2026-04-22 URL: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-embedding-2-generally-available/

Summary

Google made Gemini Embedding 2 generally available through the Gemini API and the Gemini Enterprise Agent Platform. The model is natively multimodal — handling text, image, video, and audio in a single embedding space — which enables search and reasoning across mixed-media content without requiring separate pipelines per modality. It was previously in preview and had demonstrated use cases in e-commerce discovery and video analysis.

Implications

Natively multimodal embeddings are the infrastructure layer that makes multimodal RAG (retrieval-augmented generation) practical at enterprise scale. The current state of the art requires separate embedding models per modality and a fusion layer — Gemini Embedding 2 collapses that into a single API call, which reduces pipeline complexity and latency while opening up cross-modal retrieval (find images relevant to a text query, or text relevant to a video clip) as a first-class operation.

This is a quiet but substantive move in the embedding market, which is currently dominated by OpenAI’s text-embedding-3 series and Cohere’s embed models. Neither competitor has shipped a production-ready multimodal embedding at this scope. If Google’s quality holds in independent benchmarks, Gemini Embedding 2 becomes a compelling reason to route enterprise search infrastructure through Google Cloud even for organizations that use other models for generation.

Cloud agent platform thread: Embeddings are the retrieval backbone of every RAG-based agent. Gemini Enterprise Agent Platform + Gemini Embedding 2 is a closed-loop story — Google owns the full stack from embedding to retrieval to generation to governance.

Watch:

• Independent MTEB benchmark results for Gemini Embedding 2 across modalities
• Pricing relative to OpenAI text-embedding-3-large and Cohere embed-v3-multilingual
• Whether video embedding capability creates traction in media and e-commerce verticals specifically