OpenAI text-embedding-3-small

A high-efficiency vector model delivering 1536-dimensional embeddings with a 5x price reduction over previous generations.

Released in January 2024, text-embedding-3-small is OpenAI's most cost-effective embedding model. It handles an 8192-token context window and outperforms the older ada-002 model on the MIRACL benchmark (increasing scores from 31.4% to 44.0%). Developers can use the dimensions parameter to shorten vectors without losing significant signal, making it ideal for high-volume RAG (Retrieval-Augmented Generation) pipelines and semantic search where latency and storage costs are critical.

https://platform.openai.com/docs/guides/embeddings