With text-embedding-3-small being 5x cheaper than text-embedding-ada-002, it will be significantly cheaper to perform vector embeddings of sentences. Furthermore, text-embedding-3-large boasts a larger dimension size of 3072 dimensions, which is a significant increase from the 1536 dimensions of text-embedding-ada-002. This helps to make the embeddings more expressive and better able to distinguish unique meanings from one another.

The smaller model text-embedding-3-small reduces the cost of Retrieval Augmented Generation (RAG) and increases viability for business use cases. The larger model text-embedding-3-large makes the embeddings more expressive and could potentially improve retrieval accuracy for RAG.

New embedding models and API updates