GoogleGemini Embedding
Gemini Embedding 2
Gemini Embedding 2 maps text, images, video, audio and PDFs into one embedding space.
What the release provides
The API accepts up to 8,192 input tokens. Output dimensions can range from 128 to 3,072; Google recommends 768, 1,536 or 3,072.Gemini Embedding 2 official API documentation
How to use it well
Practical guidance. Choose dimensions with a retrieval evaluation on the actual collection. Keep the model and vector length consistent between stored documents and incoming queries. Review cross-modal retrieval separately from text search so one strong result does not hide a weak modality.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API |
| Model identifiers | gemini-embedding-2 |
| Context window | 8,192 tokens API input limit; reserve output capacity where the service shares one window. Gemini Embedding 2 official API documentation |