Embedding cost: what does ingesting your documents cost?
Before an AI can search your documents, they are converted into embeddings. Enter how many documents you have and how often they change, and see the cost: one-time, per month and in the first year.
Example prices · all editable
What are embeddings?
An embedding is a series of numbers that describes the meaning of a text section. Similar content gets similar numbers. That way a RAG platform finds matching passages even when the question uses different words from the document.
Every section is converted once during ingestion and afterwards only when it changes. That is why embeddings are usually the smallest cost item of an AI platform. Still, it is worth looking at the numbers before you connect large collections such as drives or ticket systems.
How the tool calculates.
| Value | Formula |
|---|---|
| Tokens | documents × words × 1.33 |
| One-time | tokens ÷ 1M × price |
| Per month | one-time cost × change per month |
| First year | one-time + 12 × per month |
Example prices per 1M tokens, as of September 2026: small API model €0.02, large €0.13, local model €0. Enter your provider’s prices.
What do 5,000 documents cost?
A company with 5,000 documents of 1,500 words each has about 10 million tokens. With a small API model at €0.02 per million tokens, the first ingestion costs about €0.20. If 10 percent of documents change each month, about €0.02 a month is added. Even with a large model it stays at a few euros a year.
It only gets more expensive with very large collections, frequent changes or when whole collections are re-ingested again and again. That is why it pays to re-ingest only what changed.
What matters most for total cost is the language model. The AI cost calculator shows both together for all four RAGSuite plans.
When documents are re-ingested.
Content changed
Only changed sections are converted again.
Model changed
A new embedding model needs a new index: once for all documents.
Chunking changed
New chunk sizes mean new sections and therefore new embeddings.
How to keep embedding costs low.
Only re-ingest what changed
Connectors that detect changes save most of the cost.
Leave out what is not needed
Archives, duplicates and drafts do not belong in the index.
Run it locally
Open embedding models run on your hardware with no cost per token.
Questions about embedding costs.
How is the size of the documents estimated?
From the number of documents times the average number of words, converted at about 1.33 tokens per word. The cost is these tokens times the price per million tokens. For ongoing costs, only the share of documents that changes each month counts.
Why does a local model cost €0?
Open embedding models run on your own hardware. There is no fee per token; you only pay for compute you already run. That cost does not grow with the amount of your documents. With RAGSuite self-hosted, you can compute embeddings entirely locally.
Are the prices current?
They are example prices from September 2026, converted into euros. Providers change their prices regularly, sometimes several times a year. Choose “Custom price” and enter your provider’s current price. Nothing is fetched or stored, and every value can be changed at any time.
Are embeddings included in the RAGSuite Engine?
Yes. In RAGSuite Engine, the language model and embeddings run on the credits of your plan, which all users in a workspace share. With your own API key (BYOK) you use no credits. There is no price per question. The free AI cost calculator on this website shows roughly how high your AI usage is.
What costs more: embeddings or the language model?
Almost always the language model. Embeddings arise once per section; the language model on every question, with question, context and answer. With many questions, model costs quickly exceed embedding costs many times over. Work out both together in the AI cost calculator.
Embeddings locally or included.
Self-hosted you compute embeddings locally; in the RAGSuite Engine they are included in the price. Install the Community Edition or try the Cloud.