We’re excited to announce that our first Gemini Embedding textual content mannequin (gemini-embedding-001) is now typically accessible to builders within the Gemini API and Vertex AI.
This embedding mannequin has persistently held a prime spot on the Huge Textual content Embedding Benchmark (MTEB) Multilingual leaderboard because the experimental launch in March.
Surpassing each our earlier textual content embedding fashions and exterior choices in various duties, from retrieval to classification, gemini-embedding-001 offers a unified innovative expertise throughout domains, together with science, authorized, finance, and coding. Right here is how Gemini Embedding compares to different commercially accessible proprietary fashions:
*Legacy Google fashions are a mixture of the best scores from 3 Gemini API and VertexAI fashions: text-embedding-004, text-embedding-005, and text-multilingual-embedding-002
Extra detailed outcomes can be found in our technical report*.
Mannequin particulars
An extremely versatile mannequin, Gemini Embedding helps over 100 languages and has a 2048 most enter token size.
It additionally makes use of the Matryoshka Illustration Studying (MRL) method, which permits builders to scale the output dimensions down from the default 3072. This flexibility allows you to optimize for efficiency and storage prices to suit your particular wants. For the best high quality outcomes, we suggest utilizing 3072, 1536, or 768 output dimensions.
Fee limits and pricing
We provide each free and paid tiers within the Gemini API, so you possibly can experiment with gemini-embedding-001 for gratis, or ramp up with considerably increased limits on your manufacturing wants.
The Gemini Embedding mannequin is priced at $0.15 per 1M enter tokens.
Begin constructing with Gemini Embedding
Builders can now entry the Gemini Embedding mannequin (gemini-embedding-001) by way of the Gemini API, which you can begin working with free of charge via Google AI Studio.
It’s appropriate with the present embed_content endpoint.
from google import genai
shopper = genai.Shopper()
end result = shopper.fashions.embed_content(
mannequin="gemini-embedding-001",
contents="What's the that means of life?"
)
print(end result.embeddings)Python
To get began, try the official developer documentation and cookbooks:
If you’re utilizing the experimental gemini-embedding-exp-03-07, you received’t must re-embed your contents however it would now not be supported by the Gemini API on August 14, 2025. Legacy fashions may even be deprecated within the coming months:
- embedding-001 on August 14, 2025 and
- text-embedding-004 on January 14, 2026
We extremely suggest migrating your initiatives to our latest mannequin as early as potential.
We won’t wait to see how Gemini Embedding unlocks new use circumstances that weren’t beforehand potential. As well as, we may have help for Gemini Embedding within the Batch API quickly, which permits asynchronous processing of your knowledge for decrease prices.
Maintain a watch out for future bulletins concerning embedding fashions with even broader modalities and capabilities!
*MTEB benchmark ends in the printed paper replicate the experimental model of Gemini Embedding, launched in March 2025.
