Introduction to Gemini Embedding

Google has announced the addition of a new, experimental embedding model for text, called Gemini Embedding, to its Gemini developer API. Embedding models translate text inputs into numerical representations, known as embeddings, that capture the semantic meaning of the text. These embeddings are used in various applications, such as document retrieval and classification, due to their ability to reduce costs while improving latency.

Gemini Embedding is Google's first embedding model trained on the Gemini family of AI models. According to Google, this model has inherited Gemini's understanding of language and nuanced context, making it suitable for a wide range of uses, including finance, science, legal, search, and more.

Performance and Capabilities

Google claims that Gemini Embedding surpasses the performance of its previous state-of-the-art embedding model, text-embedding-004, and achieves competitive performance on popular embedding benchmarks. Compared to text-embedding-004, Gemini Embedding can also accept larger chunks of text and code at once and supports twice as many languages, with over 100 languages supported.

However, Google notes that Gemini Embedding is currently in an experimental phase with limited capacity and is subject to change. The company is working towards a stable, generally available release in the coming months.