Embedding 3 Large
Available nowFind more relevant information in demanding multilingual knowledge and search applications.
What it does
OpenAI’s text-embedding-3-large is the higher-capability model in its third-generation text embedding family. It represents relationships between pieces of text, helping a system retrieve the material that is relevant to a user’s question.
The official documentation positions it as OpenAI’s most capable embedding model for English and non-English tasks. Its default 3,072-dimensional vectors provide more room to represent meaning; shorter outputs are supported when you need to balance quality and storage.
Language coverage
OpenAI positions this model for English and non-English tasks, and reports multilingual retrieval results on the 18-language MIRACL benchmark. Evaluate your actual languages and scripts before selecting it.
Official language coverage ↗Where it helps
Multilingual knowledge
Retrieve useful passages from documents written in different languages.
High-value search
Evaluate difficult queries where a missed match creates extra work for customers.
Grounded assistants
Find relevant reference material for an assistant to use when answering.
When to choose it
Consider it when retrieval quality has a clear business value and you can justify the higher service price. Compare both OpenAI models on the same questions to see whether Large improves your actual results.
Larger default vectors need more index storage. Changing models or vector dimensions normally requires rebuilding the stored document vectors; do not mix unrelated vector spaces.
Official sources
Paraphrased from the provider’s documentation and original model information. Selection advice is our editorial guidance.
OpenAI · Official release ↗OpenAI · Embeddings guide ↗Reviewed October 3, 2026