AI API Lab

WORKBENCH / TEXT EMBEDDING

Gemini EmbeddingGoogle Gemini

Compare the meaning of documents and questions, or turn text into vectors.

02

Documents and questions

TEXT → SEARCH

Documents to search 4 / 20

Items separated by divider: 4 documents

Search questions 2 / 10

Items separated by divider: 2 questions

Up to 8,000 characters per item. The model input limit is 8,192tokens, so split long documents.

The first search embeds 4documents and 2questions separately. Document vectors are reused when documents, model, and dimensions match.

API usage and costs apply to your Google account.

03

Search results

AWAITING INPUT
⌬

Find documents relevant to your questions

Enter your API key and select “Search documents”.
Top documents and similarity scores appear for each question.

Documents use the retrieval-document task; questions use
the retrieval-query task automatically.

EXPLORE / HOW TO USE

Gemini embeddings and document search

Compare Gemini embedding models with your own Google API key. Rank documents by similarity to questions, inspect text vectors, and download results as JSON.

What you can try

  • Choose an embedding model and output dimensions.
  • Rank documents for each question using cosine similarity.
  • Inspect a single text vector or download search results and vectors as JSON.
  • Requests use your provider’s API quota and may incur charges. Check the key, model access, and billing if you receive an authentication or quota error.

How to get started

  1. Enter a Google API key and select the model and dimensions.
  2. Separate documents and questions with a line containing ---, or select single-text mode.
  3. Run the experiment, compare results, and download the JSON if needed.
  4. Example: add product descriptions as documents and ask which product fits a specific need. Compare the ranking using the same documents and questions.
Explore all playgrounds →API key guide →