Embedding Generation Cost Calculator

Work out embedding generation cost instantly with clear inputs, formula shown and shareable results.

About Embedding Generation Cost Calculator

Embedding is billed per input token, so the total is documents times tokens per document times the price. It is a one-off cost for a static corpus but recurs on every re-index, which is what makes changing embedding model an expensive decision. The resulting vector payload is documents times dimensions times bytes per value and is often the larger long-term cost.

Formula

Embedding cost

tokens = documents x tokens per document; cost = tokens x price / 1,000,000; vector bytes = documents x dimensions x 4

Worked example

With the default values below, Embedding Generation Cost Calculator returns One-off embedding cost of $14.00. Change any input above and every figure here is recalculated in your browser.

Inputs used

Documents or chunks to embedminimum 1: 2,000,000
Tokens per documentminimum 1: 350
Embedding price per million tokensminimum 0: 0.02
Embedding dimensionsminimum 1: 1,536

Results

One-off embedding cost: $14.00
Tokens embedded: 700,000,000
Cost per 1,000 documents: $0.0070
Resulting vector storage: 11.44 GiB

Step by step

1Values used
2Embedding cost
3One-off embedding cost= 14.00
4Tokens embedded= 700,000,000
5Cost per 1,000 documents= 0.0070
6Resulting vector storage= 12,288,000,000

Assumptions and limits

The result is derived only from the 4 inputs listed above. Anything not among them is not part of the calculation.
In particular, this calculator does not take inflation, tax, fees or charges or compounding frequency into account.
Figures are rounded for display to 4 decimal places, so totals built from several rounded parts can differ slightly from the unrounded arithmetic.

How these calculators are built and tested is described on the Methodology page. If a figure looks wrong, please report it.

Frequently Asked Questions

Do I need to re-embed when the model changes?

Yes, entirely. Vectors from different models live in incompatible spaces, so a model upgrade means a full re-index of the corpus.

Can dimensions be reduced after the fact?

Models trained with Matryoshka representation learning support truncating the vector to a shorter prefix with modest recall loss, which cuts storage without re-embedding.