What is a text embedding?

An embedding is a list of numbers, a vector, that represents the meaning of a piece of text. Texts with similar meaning map to vectors that point in similar directions, which lets you measure semantic similarity with simple math instead of matching keywords.

Which OpenAI embedding model should I use?

text-embedding-3-small is cheap and strong enough for most search and clustering tasks; text-embedding-3-large gives higher quality at higher cost and larger dimensions. Both support a dimensions parameter so you can shrink vectors to save storage and speed up search.

Why cosine similarity instead of plain distance?

Cosine similarity measures the angle between two vectors and ignores their magnitude, so it focuses on direction, which captures meaning. It is the standard choice for normalized embeddings and is what most vector databases use by default.

How do I store millions of vectors?

Use a vector database. pgvector adds vector columns and indexes to Postgres, while managed services like Pinecone, Qdrant, and Weaviate handle indexing, filtering, and scaling for you. They use approximate nearest-neighbor indexes to search large sets in milliseconds.

Can I compare embeddings from different models?

No. Vectors from different models, or even different dimension settings, live in different spaces and are not comparable. If you change your embedding model, you must re-embed your entire corpus so all vectors are consistent.

How to Create Text Embeddings with OpenAI

What embeddings are

A text embedding turns a string into a fixed-length vector of numbers that encodes its meaning. Two sentences that say the same thing in different words produce vectors pointing in nearly the same direction, even with no shared keywords. That property is the foundation of semantic search, clustering, classification, recommendation, and retrieval-augmented generation.

How to create and use them

You call the /v1/embeddings endpoint with a model — text-embedding-3-small for a cheap, capable default or text-embedding-3-large for higher quality — and get back an array of floats per input. You store each vector alongside its source text in a vector store like pgvector, Pinecone, or Qdrant. To find related items, you embed a query and compare it to stored vectors using cosine similarity, which measures the angle between two vectors: values near 1 mean very similar, near 0 mean unrelated.

The calculator below lets you paste two comma-separated vectors and see their cosine similarity, dot product, and magnitudes — the exact math a vector database runs under the hood.

Tips and pitfalls

Pick one embedding model and stick with it; vectors from different models are not comparable, so changing models means re-embedding everything. Use the dimensions parameter to shrink large vectors when storage or speed matters — modern OpenAI models support this with little quality loss. Normalize vectors if your store expects it, batch many texts per request to cut cost, and always keep the original text with each vector so you can show real results, not just numbers.