Embedding Model

OpenAI Embedding 3 Small

OpenAI Embedding 3 Small is a text embedding model from OpenAI with an 8,191 token context window.

Start Building with OpenAI Embedding 3 Small View All Models

Publisher

OpenAI

TypeEmbedding

Context Window8,191 tokens

Released2024

Input$0.02/MTok

Try OpenAI Embedding 3 Small →

About OpenAI Embedding 3 Small

Efficient text embeddings for semantic search

OpenAI Embedding 3 Small (text-embedding-3-small) is a text embedding model released by OpenAI in 2024. It converts text into numerical vector representations that capture semantic meaning, enabling applications such as semantic search, clustering, classification, and retrieval-augmented generation. The model accepts up to 8,191 tokens of input text and produces fixed-length vector embeddings that can be stored and queried against a vector database.

This model is the smaller of OpenAI's third-generation embedding models, designed to offer a balance between embedding quality and cost efficiency. It supports a feature called dimension reduction, which allows developers to shorten the output embedding vectors without losing significant representational quality, using a technique described in the Matryoshka Representation Learning paper. It is well suited for applications that require large-scale embedding generation, such as document retrieval systems, recommendation engines, and similarity search pipelines.

Capabilities

What OpenAI Embedding 3 Small supports

Text Embedding

Converts input text into dense numerical vectors that encode semantic meaning. Supports up to 8,191 tokens of input per request.

Dimension Reduction

Allows output embedding vectors to be shortened to a smaller dimension without retraining, using Matryoshka Representation Learning techniques.

Semantic Search

Enables similarity-based document retrieval by comparing embedding vectors, making it suitable for search and retrieval-augmented generation pipelines.

Text Classification

Embedding vectors can be used as input features for downstream classification tasks without additional fine-tuning of the model itself.

Clustering

Groups semantically similar texts by comparing their vector representations, useful for topic modeling and content organization at scale.

Ready to build with OpenAI Embedding 3 Small?

Get Started Free

FAQ

Common questions about OpenAI Embedding 3 Small

What is the context window for OpenAI Embedding 3 Small?

The model supports a maximum input of 8,191 tokens per request. Text exceeding this limit must be chunked before embedding.

What does this model output?

It outputs a fixed-length numerical vector (embedding) representing the semantic content of the input text. The default output dimension can also be reduced to a smaller size using the model's built-in dimension reduction feature.

What is the difference between text-embedding-3-small and text-embedding-3-large?

text-embedding-3-small is the smaller, lower-cost variant in OpenAI's third-generation embedding model family. text-embedding-3-large produces higher-dimensional embeddings and generally scores higher on benchmarks, but at a higher cost per token.

What is the knowledge cutoff for this model?

As an embedding model, text-embedding-3-small does not have a knowledge cutoff in the traditional sense. It does not generate text or answer questions; it encodes input text into vectors regardless of when that text was written.

What are the pricing details for this model?

Pricing information was not included in the available metadata. Current pricing can be found on OpenAI's official pricing page at platform.openai.com/docs/pricing.

Can this model process images or other non-text inputs?

No. text-embedding-3-small is a text-only embedding model. It does not support image, audio, or video inputs.

Resources