text-embedding-3-large

text-embedding-3-large is OpenAI's most capable embedding model, according to its model page, "for both english and non-english tasks". It converts text into vectors for search, clustering, recommendations, anomaly detection and classification. It belongs to OpenAI's third-generation embedding line, which also includes text-embedding-3-small, and succeeds text-embedding-ada-002.

Status

Live. No deprecation notice appears on the model page, which lists a single snapshot named text-embedding-3-large.

Key Features

• 3072 default dimensions, according to OpenAI's embeddings guide.
• Shortenable vectors: the dimensions parameter removes trailing values while keeping the embedding's useful properties. The guide's example says a text-embedding-3-large embedding shortened to 256 dimensions outperforms text-embedding-ada-002 at 1536 dimensions on MTEB. It suggests 1024 for vector stores with size limits, and says to L2-normalize if you truncate manually.
• 8192-token maximum input, per the guide.
• MTEB: the guide lists 64.6%, against 62.3% for text-embedding-3-small and 61.0% for text-embedding-ada-002. These are OpenAI's own figures.
• Batch support: the model page lists the Batch endpoint as supported.

Use Cases

• Semantic search and retrieval-augmented generation (RAG).
• Clustering, deduplication and recommendations.
• Classification and anomaly detection on text.

Limitation: it is a hosted model, so your text goes to OpenAI's API, and larger vectors cost more to store.

Pricing

OpenAI's model page lists $0.13 per 1M tokens, checked on 2026-10-29. The embeddings guide describes billing per input token and gives about 9,615 pages per US dollar (assuming roughly 800 tokens per page). The model page shows no separate Batch price.

Getting Started

1. Call the embeddings endpoint with model set to text-embedding-3-large.
2. Optionally pass dimensions (for example 1024) to shrink the vectors.
3. Store the vectors in your index and embed queries with the same model and dimension setting.

Frequently Asked Questions

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

3072 by default, and you can shorten them with the dimensions parameter.

What is the maximum input length?

8192 tokens, per OpenAI's embeddings guide.

How much does it cost?

$0.13 per 1M tokens on OpenAI's model page, as checked on 2026-10-29.

Is it better than text-embedding-3-small?

OpenAI lists 64.6% on MTEB for 3-large against 62.3% for 3-small. 3-large is the more capable model, and 3-small is the cheaper one.

Alternatives

• text-embedding-3-small: OpenAI's cheaper third-generation embedding model.
• BGE-M3: an open multilingual embedding model you can self-host.
• Voyage 3 Large: Voyage AI's large embedding model.

Conclusion

text-embedding-3-large is OpenAI's top embedding model, with flexible dimensions, 8192-token inputs and a clear per-token price. Read the model page and embeddings guide, and compare it with open or competing embedders on your own corpus.