wzlxp.top builder tools RAG calculator
Estimate retrieval-augmented generation cost from document chunks, embedding volume, retrieved context, model tokens, and monthly query volume. Start with OpenAI embedding and generation prices, then refine with production logs later.
Average words per document is shown for planning context. The estimate uses chunk count and average tokens per chunk so you can match your real splitter settings.
Generation pricing checked 2026-05-09. GPT-5.4 mini: input $0.75/1M, output $4.5/1M tokens.
Embedding pricing checked 2026-05-10. text-embedding-3-small: input $0.02/1M tokens. Re-indexing: Monthly.
Cost insight: the largest cost driver is generated answers. Cutting answer tokens by 50% would save about $23.63 per month.
Estimated planning result only. Prices can change, and provider bills may include taxes, minimums, feature-specific charges, or usage adjustments. Verify production spend in the official provider dashboard.
How to use
Enter document count, chunks per document, average tokens per chunk, and how often the index is refreshed.
Set retrieved chunks, prompt overhead, output tokens, and expected query volume so recurring generation cost is visible.
Review monthly indexing cost separately from monthly query cost before changing chunking, retrieval, or model choices.
RAG cost guide
A useful RAG estimate should not collapse everything into one token number. Indexing happens when content changes; query cost repeats every time users ask questions.
Documents are converted into chunks, then chunk count is multiplied by average tokens per chunk to estimate embedding volume.
Indexing tokens are multiplied by the selected embedding model price. The monthly version depends on whether you re-index never, monthly, weekly, or daily.
Retrieved chunks become LLM input context, prompt overhead is added, output tokens are priced separately, and the result is multiplied by query volume.
A help center with moderate document count, monthly refreshes, and hundreds of daily customer questions.
A larger private documentation corpus with frequent updates and higher retrieved context per answer.
A high-traffic public support experience where query cost dominates after the first embedding pass.
References and assumptions
The first version uses manually maintained OpenAI embedding prices and the site's shared OpenAI generation model prices. Exact provider bills can differ because of taxes, discounts, model changes, caching, retries, and provider-specific feature charges.
This page does not upload documents or call provider APIs. It estimates RAG cost from numeric workload assumptions so teams can compare chunking, retrieval depth, re-indexing frequency, and query volume before implementation.
FAQ
It estimates indexing tokens, embedding cost, retrieved context tokens, generation input and output cost, per-query cost, monthly query cost, monthly re-indexing cost, and annual RAG spend.
Indexing and embedding usually happen when documents are added or refreshed, while query cost repeats every time a user asks a question. Keeping them separate makes scaling decisions clearer.
No. It uses only numeric planning inputs in the browser. It does not upload documents, ask for API keys, or call provider APIs.
The first version includes manually maintained OpenAI embedding options. Other provider-specific embedding prices can be added after the core RAG workflow is stable.