RAG Cost Calculator
Get an estimate of what a RAG system costs to build and run, and compare two setups side by side: plain RAG, Graph RAG or sending the whole document. All figures are estimates. Nothing is uploaded.
- 1Enter your documents and questions per day
- 2Choose two setups to compare
- 3Read the ratio, then export
Your workload
Setup A
Setup B
Assumptions (change if you know better)
Estimate only. Worked out from averages and public list prices; your real bill will differ.
| Estimated cost | A | B | B ÷ A |
|---|
How each estimate is worked out
What it is for
- Estimate the monthly cost of a RAG system before building it.
- Compare two approaches on the same workload and see how many times more one costs than the other.
- See which line is the largest: embedding, storage, building a graph or the answers themselves.
- Export the numbers and the working as a CSV or a short summary to share with a team.
How it works
A RAG system has two kinds of cost. Setup happens once: every passage is turned into a vector, and for Graph RAG a language model also reads every passage to build the graph. Running costs repeat every month: storing the vectors, re-indexing documents that changed, and above all the model reading the retrieved passages and writing an answer for each question.
The calculator works these out from a few inputs. Pages are converted to words and tokens, the document is divided into passages of the size you choose, and each line is the number of tokens multiplied by the price per million tokens of the model you picked. The working for every line is shown under the table and is included in the CSV export.
The comparison column divides setup B by setup A. A value of 3 means B costs three times as much. Try plain RAG against Graph RAG to see what the graph adds, a small answer model against a large one, or RAG against sending the whole document with every question.
These are estimates for planning. The figures for Graph RAG depend heavily on the assumptions about how much the model reads and writes while building the graph; they are listed under Assumptions and can be changed.
Questions
What does RAG cost per month?
For most systems the answers dominate: the model reads a few thousand tokens of retrieved text for each question and writes a few hundred. Embedding and vector storage are usually a small fraction of the total unless the document set is very large.
How much more does Graph RAG cost than plain RAG?
Building the graph is the big difference, because a language model has to read every passage. It is paid once, and again for any documents that change. Per question, local graph search costs a little more than plain RAG, while global search can cost many times more because it reads every summary.
Is RAG cheaper than putting the whole document in the prompt?
Almost always once the document is more than a few pages, because the whole document is paid for again with every question. Choose "Whole document in every prompt" for one setup to see the ratio. Prompt caching, which some providers offer, reduces that cost and is not included here.
What is not included?
Reranking models, prompt caching and batch discounts, free tiers, plan minimums other than the fixed fee you enter, read and write charges of hosted vector databases, servers, and the time of the people building it.
Where do the prices come from?
From public price lists on the dates shown under the table. Prices change often. Choose "Custom" for any model to enter your own price.
How accurate is it?
It is a planning estimate. Token counts are worked out from word counts with an average ratio, and real documents and questions vary. Treat the ratio between two setups as more reliable than the absolute amounts.
Is my text uploaded?
No. Everything runs in your browser tab. Your documents and questions never leave your device.
More
Try RAG on your own document · Chunking Visualizer · Token Counter & Cost Estimator · Compare Two RAG Setups · Graph RAG Playground · What is RAG? · How to choose a chunk size for RAG · What are embeddings? · What are tokens, and what do they cost? · Hybrid search: words plus meaning · How to test RAG retrieval · Seven common RAG mistakes · What is GraphRAG?