Applications
RAG monthly cost estimator
Monthly cost of a retrieval-augmented answer flow: query embedding, retrieved chunks in the prompt, and the generated answer.
Prices read from the vendor pages on 2026-10-20. Arithmetic on list prices, not advice or a quote.
Saving and exports need Norwell Tools Pro, $9 a month. See what is in it. Everything above is free.
How it works
In a RAG flow the retrieved chunks are the bulk of the input tokens, and they differ per query so they cannot be cached. The system prompt can. Embedding the question is a rounding error next to generation.
The formula
prompt = system + question + chunks * tokens_per_chunk; generation = llm-api-cost(prompt, answer) with the system share cached; embedding = question_tokens * embed_priceWorked example
With the inputs the page opens on:
- queries per month: 50000
- chunks per query: 5
- tokens per chunk: 400 tokens
- system prompt tokens: 600 tokens
- question tokens: 60 tokens
- answer tokens: 300 tokens
- generation model: claude-sonnet-5-5
- embedding model: text-embedding-3-small
- system prompt cached: true
gives
- monthly cost: $359.06
- cost per query: $0.00718
- generation cost: $359
- embedding cost: $0.0600
- prompt tokens per query: 2,660 tokens
Sources
- https://developers.openai.com/api/docs/pricing (read 2026-10-20)
- https://platform.claude.com/docs/en/about-claude/pricing (read 2026-10-20)
- https://ai.google.dev/gemini-api/docs/pricing (read 2026-10-20)
Prices read from each vendor's own pricing page on 2026-10-20; the page may have changed since. This is arithmetic on published list prices, not advice or a quote. Check the vendor's page before committing a budget.
Questions
Where is the vector database?
Related tools
Monthly cost of a chat assistant where every turn re-sends the conversation so far.
Cost of an agent loop where each step re-reads the growing transcript: steps, tool results and output per step.
AI cost per active user a month, gross margin at your price, and the break-even price for an AI feature.
One-time and monthly cost of embedding a document set, by model.