Embedding Cost Calculator
Work out what it costs to turn your documents into embeddings for search or RAG, keep them up to date and embed search queries, at the price per million tokens you pay.
About 1,333 tokens per document (0.75 words a token).
Example prices only, to show a range — check your provider's pricing page for the real rate.
These are embedded again every month.
Each query is embedded too, to search your vectors.
Rules of thumb for English text. Code and other languages often use more tokens.
Each chunk becomes one vector.
Overlap tokens are embedded twice.
On document embedding only, not queries.
A short question is 10–30 tokens.
e.g. 1 if you switch model once a year.
How your embedding cost is worked out
Count the tokens, add the overlap, multiply by your price.
- 1
Count the tokens
100,000 × 1,333= 133.3M tokens
- 2
Chunk with overlap
3 chunks a doc+ 10.2M overlap= 143.5M tokens
- 3
One-off cost
143.5M × $0.02 ÷ 1M= $2.87
- 4
Each month
5% redo $0.144+ queries $0.04= $0.184 a month
Read more: the formula, a worked example and size rules
An embedding model turns each chunk of your text into a list of numbers (a vector) so it can be searched by meaning. You pay for every token the model reads. So the cost is mostly a count: how many tokens you send × your price per million tokens. You pay once to embed everything, then a little each month to embed new or changed documents and every search query.
The formula
tokens per document = words ÷ words per token (or pages × words per page ÷ words per token, or characters ÷ characters per token), rounded
overlap tokens = chunk size × overlap %, rounded
chunks per document = (tokens − overlap) ÷ (chunk size − overlap), rounded up (1 if the document fits in one chunk)
tokens embedded = documents × (tokens per document + (chunks − 1) × overlap)
one-off cost = tokens embedded ÷ 1,000,000 × price per 1M × (1 − batch discount %)
re-embedding a month = tokens embedded × monthly documents % ÷ 1,000,000 × price × (1 − batch discount %)
queries a month = queries × tokens per query ÷ 1,000,000 × price
monthly cost = re-embedding + queries
first year = one-off + 12 × monthly + full re-embeds × one-off
vector storage = documents × chunks per document × dimensions × bytes per number
Worked example
Take 100,000 documents of 1,000 words at a price of 0.02 per 1M tokens. One document is 1,000 ÷ 0.75 = 1,333.3, so 1,333 tokens, and all of them hold 100,000 × 1,333 = 133,300,000 tokens.
Cut into 512-token chunks with 10% overlap: 512 × 10% = 51.2, so 51 tokens overlap. Each document makes (1,333 − 51) ÷ (512 − 51) = 1,282 ÷ 461 = 2.78, rounded up to 3 chunks, and the two repeats add 2 × 51 = 102 tokens: 1,435 tokens a document. In total 100,000 × 1,435 = 143,500,000 tokens (10,200,000 of them from overlap) and 300,000 chunks.
One-off cost: 143.5 × 0.02 = 2.87. Each month, 5% of documents change: 143,500,000 × 5% = 7,175,000 tokens, costing 7.175 × 0.02 = 0.1435. 100,000 queries of 20 tokens are 2,000,000 tokens, costing 2 × 0.02 = 0.04. So the monthly cost is 0.1435 + 0.04 = 0.1835, and the first year is 2.87 + 12 × 0.1835 = 5.072.
Storage: 300,000 chunks × 1,536 dimensions × 4 bytes (float32) = 1,843,200,000 bytes = 1.843 GB. The same vectors in int8 are a quarter of that, 0.4608 GB.
Size rules used
| You enter | Tokens | Rule (editable in Advanced) |
|---|---|---|
| 1,000 words | 1,333 | 0.75 words per token |
| 1 page | 667 | 500 words per page |
| 4,000 characters | 1,000 | 4 characters per token |
- Embedding is usually the small part of a search or RAG bill. The vector database, and the chat model that writes answers, often cost far more each month. See the vector database size and LLM API cost calculators for those.
- Smaller chunks make more vectors. The token cost changes only by the overlap, but storage and search work grow with every chunk.
- Switching model means re-embedding everything, because vectors from different models can't be compared. Put the number of switches you expect in "Full re-embeds a year".
- Count your own tokens on a sample document; tokenisers differ between models and languages.
No provider prices are built in: enter the rate you pay. The example price buttons only show a range and aren't any provider's current price. Rules checked 2026-10-15.
Common questions
How much does it cost to embed 1 million tokens?
Exactly your provider's price per million tokens. Embedding prices are quoted per 1M input tokens, so 1M tokens at a price of 0.02 costs 0.02, and 143.5M tokens costs 143.5 × 0.02 = 2.87. Check your provider's pricing page for the current rate.
How many tokens are in my documents?
A common rule for English is about 0.75 words per token (so 1,000 words ≈ 1,333 tokens), about 4 characters per token and about 500 words per page. Code, numbers and other languages often use more tokens. Use a token counter on a sample of your own text for a closer figure, then change the ratio in Advanced.
Does chunk overlap cost more?
Yes. With overlap, the end of each chunk is repeated at the start of the next, so those tokens are embedded twice. A 1,333-token document in 512-token chunks with 10% (51-token) overlap makes 3 chunks and 1,333 + 2 × 51 = 1,435 tokens, about 7.7% more than without overlap.
Do I pay to embed search queries too?
Yes. Each search query is turned into a vector with the same model so it can be compared with your documents. Queries are short (often 10–30 tokens), so even 100,000 queries a month of 20 tokens is only 2M tokens.
What does a batch discount apply to?
Some providers charge less for jobs sent in a batch that can take hours to finish. That suits embedding your documents, but not search queries, which need an answer straight away. This calculator applies your batch discount to the document embedding (one-off and monthly updates) only.
How much storage do the vectors need?
One vector per chunk. Each vector is dimensions × bytes per number: 1,536 dimensions in float32 (4 bytes) is 6,144 bytes. 300,000 chunks take 300,000 × 6,144 = 1,843,200,000 bytes, about 1.84 GB, before any index or metadata.
Related calculators
Made by Zulu Calculator, part of AlphaZulu LLC. Formulas are shown in full and rule figures are checked against official sources; see how we build calculators.