Gemini Embedding 2
Awaiting activationSearch across text, images, audio, video and PDFs in a single multilingual vector space.
What it does
Gemini Embedding 2 is Google’s first native multimodal embedding model in the Gemini API. It connects text, images, audio, video and PDFs in one shared space, enabling a text query to retrieve relevant visual or recorded content. The official guide lists an 8,192-token window and recommended 768, 1,536 and 3,072 dimensions.
Google permits up to six PNG/JPEG images, 180 seconds of MP3/WAV audio, 120 seconds of MP4/MOV video, or one PDF of up to six pages per request. Video is sampled to at most 32 frames and its audio track is not processed. The AIRuntime adapter uses inline media; this service applies a smaller 1 MiB request cap. Google remains paused on embeddings.app.
Language coverage
Google describes broad multilingual coverage for Gemini Embedding 2. The service remains awaiting activation; language selection does not enable a pending model.
Official language coverage ↗Content types
text · image · audio · video · PDF documents
Search across text, images, audio, video and PDFs in one shared space. Video audio tracks are not processed.
AIRuntime media adapter implemented; commercial media activation is pending. The media API awaits commercial activation.
Where it helps
Cross-media knowledge
Find evidence across text, charts, page images and PDFs.
Audio discovery
Match a text query with relevant short recordings.
Visual clip search
Retrieve visual scenes; supply separate audio when the soundtrack is needed.
When to choose it
Evaluate it when multiple content types should share one searchable representation. For video audio and temporal relationships, also compare a dedicated video model.
Google activation remains pending as requested. Text-byte pricing does not apply to media. Official rates differ for text, image, audio and video; media pricing must be verified before activation.
Official sources
Paraphrased from the provider’s documentation and original model information. Selection advice is our editorial guidance.
Google · Gemini Embedding 2 capabilities ↗Google · Official modality pricing ↗Reviewed October 4, 2026