RAG Pipeline Cost Calculator

How is the total monthly cost built?

Total monthly = monthly refresh embedding cost + monthly query embedding cost + monthly LLM cost. The one-time indexing cost is displayed separately and is not added to the monthly total. Cost per query = total monthly ÷ queries per month.

How is indexing cost calculated?

Indexing cost = documents × average tokens per document ÷ 1,000 × the embedding rate. It counts documents, not overlapping chunks, so chunk overlap is not reflected.

What does the monthly refresh assume?

The code assumes 10 percent of the indexing cost recurs each month, as if a tenth of the corpus is re-embedded. It is a fixed assumption with no input, so a corpus that changes faster or slower than that needs a manual adjustment to the document count.

How are query embeddings and LLM tokens counted?

Each query is embedded as a fixed 50 tokens. LLM input per query = chunks retrieved × tokens per chunk + 50, and output per query = the response tokens you enter. Both are priced at the chosen model’s input and output rates.

Which embedding and LLM prices are built in?

Two embedding entries and four LLM entries, listed in the table on this page. They are figures built into this tool, stored per 1,000 tokens. Providers revise prices and retire models, and the LLM entries carry no version numbers, so check current pricing pages.

What does the calculator not include?

Vector database hosting, reranking calls, conversation history, a system prompt beyond the 50 question tokens, query rewriting, chunk overlap, caching and retries are not modelled.

Why does the LLM line dominate the total?

With the default settings it is $19.87 of a $19.93 monthly total, because every query sends several chunks to the LLM and generates a response, while embeddings are priced at a much lower rate. Your own ratio depends on chunks retrieved, chunk size, response length and the model chosen.

How can I model a larger or smaller knowledge base?

Change documents and average tokens per document for the one-time indexing cost, and change queries, chunks and chunk size for the monthly lines. The monthly refresh scales with the indexing cost, so it moves with those two inputs.