Google launches a new text embedding model based on Gemini.

Google has introduced a new experimental embedding model for text, called Gemini Embedding, in its Gemini developers' API. Embedding models convert text inputs, such as words and phrases, into numerical representations that capture the semantic meaning of the content. These representations are used in various applications, such as document retrieval and classification, as they help reduce costs and improve latency. Companies like Amazon, Cohere, and OpenAI also offer embedding models through their APIs.

Although Google had previously provided embedding models, Gemini Embedding is the first one specifically trained on the Gemini family of AI models. According to the company, "trained on the Gemini model, this embedding model has inherited Gemini's nuanced understanding of language and context, making it applicable to a wide variety of use cases." Additionally, Google claims that this new model outperforms its previous state-of-the-art model, text-embedding-004, and achieves competitive results on popular benchmarks for embedding models.

Gemini Embedding also has the ability to process larger blocks of text and code compared to text-embedding-004 and supports double the number of languages, exceeding 100. However, the company clarifies that it is in an "experimental phase" with limited capacity and is subject to modifications. In its announcement, Google stated, "we are working towards a stable release available to everyone in the coming months."