OpenAI's text-embedding-3-large and text-embedding-3-small are the industry-standard embedding models. They're the default choice for most RAG systems simply because everyone already has an OpenAI API key. text-embedding-3-large is a 3072-dimensional model that can be shortened to any lower dimension (e.g., 1024, 512) via Matryoshka truncation without significant quality loss.
Call the /v1/embeddings endpoint with a text input, get back a dense vector. Supports up to 8191 tokens per request. The Matryoshka property means you can request a smaller dimension via the `dimensions` parameter to save storage in your vector database — useful when you have tens of millions of vectors.
text-embedding-3-small: $0.02 per M tokens. text-embedding-3-large: $0.13 per M tokens. Batch API offers 50% discount.
Default choice for RAG applications — thousands of production systems. The 'nobody ever got fired for picking OpenAI embeddings' of the embedding world.