Gemini Embedding 2Multimodal Embedding Model

Google's first natively multimodal embedding model — text, images, video, audio, and documents in a single unified space.

Text Embeddings

Up to 8192 input tokens across 100+ languages with configurable dimensions from 128 to 3072.

Process up to 6 images per request in PNG and JPEG formats with native multimodal understanding.

Embed up to 128 seconds of video in MP4/MOV with H264, H265, AV1, and VP9 codec support.

Native audio ingestion up to 80 seconds in MP3/WAV without intermediate transcription.

Directly embed PDFs up to 6 pages long for document search and classification.

Interleaved multimodal input — combine text, images, and more in a single request for unified retrieval.

Flexible output from 128 to 3072 dimensions via MRL. Balance performance and storage costs.

Free of charge for all input types. Paid tier starts at $0.20/1M tokens for text.