OpenAI’s model documentation lists embedding prices from $0.02 to $0.13 per 1 million tokens for the models below. OpenAI says Batch processing is 50% less than synchronous processing; using that discount, the corresponding Batch rates are $0.01 to $0.065 per 1 million tokens. Prices and Batch terms were checked against OpenAI documentation on November 9, 2026. (developers.openai.com)
OpenAI embedding prices
| Model | Standard, per 1M tokens | Batch, per 1M tokens* |
|---|---|---|
text-embedding-3-small |
$0.02 | $0.01 |
text-embedding-3-large |
$0.13 | $0.065 |
text-embedding-ada-002 |
$0.10 | $0.05 |
Prices are in U.S. dollars. The model pages list the standard rates; the Batch figures are calculated using OpenAI’s stated 50% discount for Batch-eligible models. Confirm model eligibility in its current reference before submitting a Batch job. (developers.openai.com)
Estimate your embedding API cost
Use this formula:
Estimated API cost = (input tokens ÷ 1,000,000) × price per 1M tokens
For example, indexing 10,000 documents at an assumed 500 tokens each means 5 million tokens for one run:
text-embedding-3-small, standard: 5 × $0.02 = $0.10.text-embedding-3-small, Batch: 5 × $0.01 = $0.05.
These estimates cover that indexing run at the listed rates. For re-indexing, calculate each run from the tokens actually processed; updates to only part of a corpus need not cost as much as re-embedding all of it.
Standard or Batch?
Choose standard processing when your application needs a synchronous response. OpenAI’s Batch API is asynchronous, supports the /v1/embeddings endpoint, and offers a 50% discount compared with synchronous API rates. OpenAI specifies a 24-hour completion window; batches may finish sooner. Batch is therefore an option for eligible embedding work when your deadline can accommodate that window—not for a response that must be returned immediately. (developers.openai.com)
Price alone does not determine which model will work best for your application. Evaluate candidates against your own search or retrieval requirements before choosing one.
What this estimate covers
The table and examples estimate embedding API charges, not the full cost of a production system. Budget separately for relevant infrastructure, such as vector storage, and for future embedding runs. The amount for repeat runs depends on how many tokens you submit each time.