On Friday, Google introduced an experimental “embedding” model called Gemini Embedding to its Gemini developer API.
Embedding models convert textual inputs—such as words and phrases—into numerical formats known as embeddings, which encapsulate the semantic meaning of the text. These embeddings are utilized across various applications, including document retrieval and classification, largely because they enhance performance while reducing costs.
Firms like Amazon, Cohere, and OpenAI provide their own embedding models via APIs. While Google has previously offered embedding models, Gemini Embedding marks the company’s first model developed with its Gemini series of AI technologies.
“This embedding model, crafted based on the Gemini model itself, incorporates Gemini’s expertise in language and subtle context, making it suitable for a wide array of applications,” Google stated in a blog post. “We’ve engineered our model to be impressively versatile, delivering outstanding results across diverse sectors such as finance, science, law, search, and beyond.”
Google asserts that Gemini Embedding outperforms its formerly leading embedding model, text-embedding-004, and shows competitive results on prominent embedding benchmarks. Unlike text-embedding-004, Gemini Embedding is capable of processing larger text and code segments simultaneously and supports twice the number of languages—over 100 in total.
The tech giant mentions that Gemini Embedding is currently in an “experimental phase” with limited capacity and may undergo changes. “[W]e’re aiming for a stable, widely available release in the coming months,” the company noted in its blog update.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
