OpenAI Embedding Model Pricing: Token Rates and Cost Guide

OpenAI embedding model pricing runs from $0.02 to $0.13 per 1,000,000 input tokens. At Layer3Labs, we help companies budget and deploy automated workflows, where input token volume dictates monthly software spend. Standard API calls bill input text only.

There is no output charge. Unlike chat models that bill both prompt and generation tokens, embedding models generate numeric vectors without charging for the output. You pay strictly for the text you send into the Application Programming Interface (API).

Choosing a model comes down to two options. Small costs $0.02 per million tokens, while large costs $0.13 per million tokens. That price gap is substantial, yet both models remain cheap for everyday production tasks.

OpenAI Embedding Model Pricing Across Current Models

OpenAI offers two primary embedding models alongside one legacy option, with prices ranging between $0.02 and $0.13 per million input tokens. The entry option is text-embedding-3-small at $0.02 per million input tokens. It handles general retrieval work.

The flagship option is text-embedding-3-large at $0.13 per million input tokens. For deferred or asynchronous processing jobs, the Batch API cuts text-embedding-3-large down to $0.065 per million tokens. The legacy option is text-embedding-ada-002 at $0.10 per million input tokens.

OpenAI recommends the newer family for all fresh development. Prices change over time. Verify current numbers on the OpenAI API pricing page before finalizing a project budget.

• text-embedding-3-small: $0.02 per 1,000,000 input tokens, producing vector representations up to 1536 dimensions.
• text-embedding-3-large: $0.13 per 1,000,000 input tokens via the standard API, producing vectors up to 3072 dimensions.
• text-embedding-3-large (Batch API): $0.065 per 1,000,000 input tokens for deferred or asynchronous processing jobs.
• text-embedding-ada-002: $0.10 per 1,000,000 input tokens with a fixed 1536-dimension output.

text-embedding-3-large Pricing Compared to text-embedding-3-small

text-embedding-3-large pricing is 6.5 times higher than text-embedding-3-small, charging $0.13 per million tokens compared to $0.02. This cost difference reflects higher vector dimensionality and improved search quality. On published retrieval benchmarks, the large model scores higher than the small version. When high precision is necessary, that performance gap justifies the higher spend.

text-embedding-3-small remains the practical default for high-volume jobs. At $0.02 per million tokens, you can process large text archives at low cost. It delivers reliable semantic search across standard product catalogs and internal documentation.

Vector dimensions affect storage downstream. text-embedding-3-large outputs vectors up to 3072 numbers long, while small outputs up to 1536. OpenAI provides a dimensions parameter that shortens the output vector without destroying semantic relationships. Shortening large vectors cuts memory usage in downstream vector databases while retaining most retrieval performance.

Corpus Indexing Estimates with Token Math

Indexing costs depend entirely on token count rather than raw character volume. As an illustrative estimate based on the standard tokenizer rule of thumb, one English word translates to roughly 1.3 tokens. A document with 1,000 words therefore contains roughly 1,300 tokens. This arithmetic provides a baseline for sizing project expenses.

Consider an archive of 1,000 documentation pages. At 1,000 words per page, the entire repository represents approximately 1,300,000 input tokens. Processing those 1.3 million tokens with text-embedding-3-small costs roughly $0.026. Running the identical collection through text-embedding-3-large costs roughly $0.169. Both figures demonstrate that initial embedding creation is rarely the main expense in an AI pipeline.

Query traffic adds ongoing costs. Each search query must be embedded before comparing it against the stored vectors. A short user query of 50 tokens processed one million times consumes 50,000,000 tokens. On text-embedding-3-small, that query volume costs $1.00. On text-embedding-3-large, it costs $6.50.

• 1,000 words: approximately 1,300 tokens under the standard 1.3 tokens per word guideline.
• 1,000 pages (1.3M tokens) on text-embedding-3-small: roughly $0.026 as an illustrative estimate.
• 1,000 pages (1.3M tokens) on text-embedding-3-large: roughly $0.169 as an illustrative estimate.
• 1,000,000 search queries (50 tokens each) on text-embedding-3-small: $1.00 total spend.
• 1,000,000 search queries (50 tokens each) on text-embedding-3-large: $6.50 total spend.

Legacy OpenAI ada-002 Pricing and Upgrade Tradeoffs

OpenAI ada-002 pricing sits at $0.10 per million input tokens, making it five times more expensive than text-embedding-3-small. Released as an earlier generation model, text-embedding-ada-002 remains active for backward compatibility. OpenAI recommends that teams choose the text-embedding-3 series for all new projects.

The technical constraints of ada-002 are rigid. Its output dimension is fixed at 1536, with no ability to reduce vector length using native parameters. Upgrading to text-embedding-3-small lowers token costs by 80 percent while providing improved retrieval quality. For details on model migration, check the OpenAI models documentation.

Existing systems may face migration hurdles. Because vector spaces between different model families are incompatible, switching models requires re-embedding your entire database. If an established system has millions of vectors already indexed and query volume is low, keeping ada-002 can avoid re-indexing overhead until a larger architectural redesign occurs.

Embedding Model Cost per Token and System Roles

Embedding models convert text strings into numerical vectors that capture semantic meaning for search, clustering, and retrieval systems. Unlike generative chat models, an embedding model does not formulate conversational answers or write prose. It powers the retrieval stage in Retrieval-Augmented Generation (RAG) by locating relevant passages before a language model synthesizes the answer.

The embedding model cost per token is only one component of search infrastructure. Storing and querying vector points introduces index hosting expenses in dedicated vector databases. For guidance on vector storage architecture, explore our vector database for RAG guide and Pinecone evaluation.

Several other commercial embedding providers exist outside OpenAI, each with its own pricing structure and vector dimensionality. Evaluate their pricing pages independently to compare rates against OpenAI for your specific retrieval workload.

How to Choose Between Embedding Models

For most production deployments, text-embedding-3-small offers the optimal balance of low cost and solid search quality. High-stakes legal, medical, or technical retrieval tasks should spend the premium on text-embedding-3-large to maximize benchmark accuracy. If you need to embed large batch collections on a deferred, non-urgent schedule, the Batch API offers a sensible halfway point at $0.065 per million tokens.

Who this is not for: Teams building offline or edge applications with zero internet connectivity cannot use cloud Application Programming Interface (API) endpoints. They should deploy local open-source embedding models on their own infrastructure instead. Similarly, organizations with strict data residency mandates requiring on-premises data isolation must host embeddings locally.

What would change our answer: A pricing change could shift this calculus. If OpenAI adds fees for dimension adjustment, or if competitors price high-dimension models below $0.02 per million tokens, our recommendation would flip. Review your token volume, test retrieval accuracy on sample data, and check the latest OpenAI embedding model pricing on the OpenAI API pricing page before provisioning production keys.

• Default recommendation: text-embedding-3-small ($0.02 per 1M tokens) for general documentation, customer portals, and cost-sensitive applications.
• Accuracy recommendation: text-embedding-3-large ($0.13 per 1M tokens) for complex domain retrieval where precision outweighs marginal token fees.
• Batch recommendation: text-embedding-3-large Batch API ($0.065 per 1M tokens) for deferred ingestion runs.