Text Embedding 3 Large

Turn text into numerical vectors for semantic search, retrieval and recommendations. Choose the vector size to balance accuracy and storage.

Modalities: Text
Embeddings
Price: $0.13per 1M input tokens
Vector dimensions: Up to 3,072
Released: Jan 25, 2024

Overview

Search by meaning, beyond exact words

Text Embedding 3 Large converts text into vectors that represent its meaning. Your application compares those vectors to retrieve related passages, group similar documents or power recommendations. It is OpenAI’s largest embedding model, released January 25, 2024, and it returns numbers rather than a written answer.

In a retrieval workflow, embed your documents once and store the vectors alongside the original text. At query time, embed the question with the same model and dimensions, find the closest stored vectors and pass those passages to a language model as context.

Choose a vector size deliberately

The model returns 3,072 dimensions by default, and the dimensions parameter shortens them. OpenAI reports that a 256-dimension vector from this model still outperforms a full 1,536-dimension Ada 002 vector on MTEB. Size drives storage: at 4 bytes per value, a million vectors take about 12.3 GB at 3,072 dimensions and about 1 GB at 256, before index overhead. Use the same size for documents and queries.

Large or Small

Text Embedding 3 Small costs $0.02 per million tokens against $0.13 for Large. OpenAI’s published averages put Large ahead on English tasks (MTEB 64.6 against 62.3) and further ahead on multilingual retrieval (MIRACL 54.9 against 44.0). If your content spans languages or retrieval quality drives the product, start with Large. For high-volume English workloads, test Small first.

Benchmarks by OpenAI

OpenAI’s published averages for its three embedding models, at their default dimensions.

MTEB average64.6English tasks · default 3,072 dimensions MIRACL average54.9Multilingual retrieval

Gain over 3 Small+10.9MIRACL points (MTEB +2.3)

Shortest that beats Ada256Dimensions, per OpenAI

MTEB average

Score · English embedding tasks

MIRACL average

Score · multilingual retrieval

Published by the model’s creator for its own models in January 2024, not an independent leaderboard. Scores from current MTEB leaderboard versions use different task sets and are not comparable. Run a retrieval test on your own documents before switching models.

Pricing

Text Embedding 3 Large pricing in US dollars
Usage	Price	Unit
Embedding generation	$0.13	1M input tokens

Compare embedding prices

AI/ML API input rates per million tokens. Embedding models have no output charge.

Model	Input per 1M tokens	Input length
Text Embedding 3 Large	$0.13	8Ktokens
Voyage Large 2	$0.12	16Ktokens
Qwen Text Embedding v4	$0.07	8Ktokens
Text Multilingual Embedding 002	$0.025	2Ktokens
Text Embedding 3 Small	$0.02	8Ktokens

Each provider counts tokens with its own tokenizer, so the same text can bill a different number of tokens.

Estimate by corpus size

Queries are billed too, at the same rate per token.

Tokens embedded	Cost on AI/ML API
100,000 tokens	$0.013
1,000,000 tokens	$0.13
10,000,000 tokens	$1.30
100,000,000 tokens	$13.00

Quick start

View docs

Build with your agent

Give your agent the model details, code and documentation.

1Get your API key

Create API key

export AIMLAPI_KEY="8f2b4c7d9a1e6b3f…"

2Make your first request

curl 'https://api.aimlapi.com/v1/embeddings' \
  -H "Authorization: Bearer $AIMLAPI_KEY" \
  -H 'Content-Type: application/json' \
  -d '{
  "model": "text-embedding-3-large",
  "input": "A practical guide to building semantic search."
}'

import os
import requests

headers = {"Authorization": "Bearer " + os.environ["AIMLAPI_KEY"]}
payload = {
    "model": "text-embedding-3-large",
    "input": "A practical guide to building semantic search."
}
response = requests.post(
    "https://api.aimlapi.com/v1/embeddings",
    headers=headers,
    json=payload,
    timeout=120,
)
response.raise_for_status()
data = response.json()
embedding = data["data"][0]["embedding"]
print(len(embedding), embedding[:5])

async function main() {
  const headers = {
    Authorization: "Bearer " + process.env.AIMLAPI_KEY,
    "Content-Type": "application/json",
  };
  const response = await fetch(
    "https://api.aimlapi.com/v1/embeddings",
    {
      method: "POST",
      headers,
      body: JSON.stringify({
        "model": "text-embedding-3-large",
        "input": "A practical guide to building semantic search."
      }),
    },
  );
  if (!response.ok) throw new Error(await response.text());
  const data = await response.json();
  const embedding = data.data[0].embedding;
  console.log(embedding.length, embedding.slice(0, 5));
}

main().catch(error => {
  console.error(error);
  process.exitCode = 1;
});

All posts

What is Retrieval Augmented Generation (RAG)?
Basic KnowledgeJuly 7, 2025
Hidden Embedding Models You Have Been Missing Out On
AI NewsAugust 26, 2024
What Is an LLM API Gateway for Production AI Apps?
Basic KnowledgeSeptember 17, 2026

Frequently asked questions

Does this model write responses?

No. It returns one numeric vector per input. Pair it with a language model if you need a written answer from the retrieved passages.

Can I reduce the vector size?

Yes. Set dimensions to shorten the default 3,072-dimension output. OpenAI reports that even 256 dimensions outperform a full Ada 002 vector on MTEB.

Can I mix vectors from different models or sizes?

No. Vectors are only comparable when they come from the same model with the same dimensions. If you change either, embed the collection again.

How much text fits in one request?

Up to 8,192 tokens per input, according to OpenAI. Split longer documents into passages before embedding them.

How is it billed?

AI/ML API charges $0.13 per million input tokens, and vectors are not billed as output. Embedding 10 million tokens costs $1.30.

How good is it?

OpenAI published an MTEB average of 64.6 and a MIRACL average of 54.9 at launch in January 2024, ahead of its Small and Ada 002 models. These are the creator’s own results and are not comparable with current MTEB leaderboard versions.

Should I use Large or Small?

Small costs $0.02 per million tokens and scores slightly lower on English tasks. Large is clearly stronger on multilingual retrieval. Test both on a sample of your own queries.

Start building with Text Embedding 3 Large

Get API key

API

API key, code and agent prompt for Text Embedding 3 Large.

Build with your agent

Give your agent the model details, code and documentation.

1Get your API key

Create API key

export AIMLAPI_KEY="8f2b4c7d9a1e6b3f…"

2Make your first request

curl 'https://api.aimlapi.com/v1/embeddings' \
  -H "Authorization: Bearer $AIMLAPI_KEY" \
  -H 'Content-Type: application/json' \
  -d '{
  "model": "text-embedding-3-large",
  "input": "A practical guide to building semantic search."
}'

import os
import requests

headers = {"Authorization": "Bearer " + os.environ["AIMLAPI_KEY"]}
payload = {
    "model": "text-embedding-3-large",
    "input": "A practical guide to building semantic search."
}
response = requests.post(
    "https://api.aimlapi.com/v1/embeddings",
    headers=headers,
    json=payload,
    timeout=120,
)
response.raise_for_status()
data = response.json()
embedding = data["data"][0]["embedding"]
print(len(embedding), embedding[:5])

async function main() {
  const headers = {
    Authorization: "Bearer " + process.env.AIMLAPI_KEY,
    "Content-Type": "application/json",
  };
  const response = await fetch(
    "https://api.aimlapi.com/v1/embeddings",
    {
      method: "POST",
      headers,
      body: JSON.stringify({
        "model": "text-embedding-3-large",
        "input": "A practical guide to building semantic search."
      }),
    },
  );
  if (!response.ok) throw new Error(await response.text());
  const data = await response.json();
  const embedding = data.data[0].embedding;
  console.log(embedding.length, embedding.slice(0, 5));
}

main().catch(error => {
  console.error(error);
  process.exitCode = 1;
});

Docs for AI agents Plain-text index · llms.txt API documentation API reference and examples

Text Embedding 3 Large

Overview

Search by meaning, beyond exact words

Choose a vector size deliberately

Large or Small

MTEB average

MIRACL average

Pricing

Compare embedding prices

Estimate by corpus size

Quick start

Build with your agent

1Get your API key

2Make your first request

Related models

Text Embedding 3 Small

Text Embedding Ada 002

Qwen Text Embedding v4

Voyage Large 2

Text Multilingual Embedding 002

Related articles

What is Retrieval Augmented Generation (RAG)?

Hidden Embedding Models You Have Been Missing Out On

What Is an LLM API Gateway for Production AI Apps?

Frequently asked questions

Start building with Text Embedding 3 Large