Source: OpenAI Vector embeddings guide

OpenAI's API guide to text-embedding-3-small/-large: dimensions, input limits, shortening via the dimensions parameter, normalization and pricing.

Contents

Summary

The guide documents OpenAI’s third-generation embedding models and how to use them for search, clustering, recommendations and classification. As of November 2026 it still lists only the -3 models and the legacy ada-002, with no newer embedding model.

Key claims

  • text-embedding-3-small: 1536 dims by default, 8,192 max input tokens, MTEB average 62.3%; $0.02 per 1M tokens.
  • text-embedding-3-large: 3072 dims by default, 8,192 max input tokens, MTEB 64.6%; $0.13 per 1M tokens.
  • text-embedding-ada-002: legacy, MTEB 61.0%; $0.10 per 1M tokens.
  • The dimensions parameter shortens embeddings; -3-large cut to 256 dims still beats ada-002 at 1536 on MTEB. Manual truncation requires re-normalizing.
  • Embeddings are normalized to length 1, so cosine similarity equals a dot product and gives the same ranking as Euclidean distance.
  • Tokenizer is cl100k_base; the models have no knowledge of events after September 2021.

Feeds

Embeddings

Suggest changes →