Back to OpenAI
OpenAI /

text-embedding-3-large
text-embedding-3-large

API OnlyEmbeddingtextUpdated January 25, 2024

Text Embedding 3 Large: OpenAI's Best Embedding Model

Model Overview

text-embedding-3-large is OpenAI's most capable text embedding model, released in January 2024. It produces 3072-dimensional embeddings with significantly improved performance over the previous Ada-002 embedding model — achieving 54.9% on MTEB (Multilingual Text Embedding Benchmark) versus Ada-002's 31.4%. It supports shortening embeddings to reduce storage/computation costs without significant performance loss.


✨ Key Features

Specification Table
FeatureDescription
3072 DimensionsHigh-dimensional embeddings for maximum semantic richness
Flexible DimensionsSupports reducing to any dimension ≤3072 via dimensions parameter
MTEB Performance54.9% on MTEB — 75% improvement over text-embedding-ada-002
MultilingualStrong multilingual embedding performance across many languages
Cost-EffectiveSignificantly cheaper than Ada-002 per token

📊 Benchmarks

Specification Table
Benchmarktext-embedding-3-largetext-embedding-ada-002
MTEB Overall54.9%31.4%
Avg Retrieval (BEIR)54.9%49.3%

🔗 Resources

Specification Table

📜 License & Access

Proprietary — Available via OpenAI API.

Key Features

Supports Matryoshka Representation Learning for truncated output dimensions

Feature 01

8,192 token context window

Feature 02

Highly cost-efficient API pricing

Feature 03

Optimized for vector database search and RAG

Feature 04

You might also want to compare

Verified Sources

Tags

embeddingsearchapi-only

Model Specs

api-only

Parameters

Undisclosed

Context Window

8192

License

Proprietary

Deployment

api-only

Resources & Links

Lineage

Curator Notes

Features Matryoshka learning allowing vectors to be truncated without severe performance drop.

Compare Specs

Compare parameters, context windows, modalities, and benchmark scores of this model side-by-side with others.

Compare Model