text-embedding-3-small

text-embedding-3-small is an OpenAI embedding model, described on its model page as an improved, more performant version of the ada embedding model. The embeddings guide gives a default output dimension of 1536 and a maximum input of 8192 tokens, and lets you shorten vectors with the dimensions parameter. It is not listed as deprecated: OpenAI's deprecations page mentions it only as the recommended replacement for first-generation embedding models that shut down on January 4, 2024.

Key features

• 1536 dimensions by default, with a dimensions parameter to shorten vectors; per the guide, normalize the vector yourself if you shorten it manually.
• 8192-token input limit per the guide.
• MTEB: the guide reports a 62.3% MTEB score for this model.
• Price: $0.02 per 1M tokens on the model page.
• Batch endpoint is supported, though no batch rate is listed on the model page.

Use cases

• Semantic search and retrieval-augmented generation over documents.
• Clustering, deduplication and recommendations where low cost matters.
• Storage-aware setups that shorten vectors with dimensions.

Pricing and availability

Price: $0.02 per 1M tokens, per OpenAI's model page, checked on 2026-10-29. Billing is based on input tokens. Status: not deprecated on the deprecations page. Vectors from different embedding models are not comparable, so switching models means re-embedding your data.

Getting started

1. Call the Embeddings API with model set to text-embedding-3-small.
2. Keep inputs under 8192 tokens, splitting long documents into chunks.
3. Optionally set dimensions to shorten vectors, and normalize if you shorten them yourself.
4. Store the model name with your vectors, so a future migration is easy to plan.

FAQ

How many dimensions does text-embedding-3-small return?

1536 by default. You can shorten vectors with the dimensions parameter.

What is the input limit?

8192 tokens, per OpenAI's embeddings guide.

How much does it cost?

$0.02 per 1M tokens, per the OpenAI model page.

Is it deprecated?

Not on OpenAI's deprecations page as of 2026-10-29.

What is the larger option?

See text-embedding-3-large.

Alternatives

• text-embedding-3-large: OpenAI's larger embedding model.
• text-embedding-ada-002: the earlier ada embedding model.
• BGE-M3: an open multilingual embedding model.

Conclusion

text-embedding-3-small remains a low-cost, documented embedding option with flexible dimensions and no deprecation notice. See the model page, the embeddings guide and the models category.