Estimate document Q&A AI cost before it adds up

Enter how many questions your document Q&A assistant answers each month to get a Low / Likely / High monthly API cost estimate. Everything runs locally in your browser — no data is sent to any server.

  • No login
  • No credit card
  • No API calls
  • Estimate only

Configure your document Q&A

Estimated monthly cost

This estimates the model-inference cost of answering a question once relevant text has already been retrieved. It does not include document ingestion, OCR, embeddings, vector-database hosting, or retrieval infrastructure — those are billed separately by whatever tools you use to build the retrieval step.

Likely monthly cost$1.01Estimated range: $0.6450$1.50
Cost per question
Low$0.001290Likely$0.002025High$0.003000
Retrieved context / question
Low300Likely500High700
Input tokens / question
Low520Likely900High1.3K
Output tokens / question
Low200Likely300High450
Total monthly tokens
Low360KLikely600KHigh875K

This is an estimate, not a billing guarantee. Actual costs depend on your provider's exact tokeniser, real retrieval size, and your billing tier.

Understand & trust these numbers Methodology What are AI tokens? Why bills surprise you Document summary calculator

How this estimate works

Input — you send
  • Instruction / system prompt
  • The user's question
  • Retrieved document context
Model

Reads the question and the retrieved passages, then writes an answer.

Output — you get

An answer grounded in the retrieved text.

"RAG" (retrieval-augmented generation) simply means: find the most relevant pieces of your documents for a question, then ask the model to answer using only that retrieved text. This calculator starts counting cost after that retrieval step has already happened.

What's not included: retrieval infrastructure

This calculator estimates the model-inference cost of answering a question once relevant text has already been retrieved — it does not include document ingestion, OCR, embeddings, vector-database hosting or search, reranking, or any other retrieval infrastructure.

Those steps are billed separately by whatever embedding model, vector database, and retrieval service you use to build the retrieval pipeline, and are not part of this estimate.

Low / Likely / High explained

Low

Focused retrieval, short question, short answer.

Likely

Standard retrieval depth and typical question/answer length. Plan against this number.

High

Broad retrieval, long question, long answer. Budget against this.

Cut your cost

1
Retrieve less. A narrower, more relevant retrieval window is usually the single biggest cost lever.
2
Keep the instruction prompt tight. It repeats on every single question.
3
Match model to task. Routine document Q&A rarely needs a frontier-tier model.

Example monthly costs

These worked examples use the same default assumptions as the calculator above — medium-length questions and answers, standard retrieval depth, a standard instruction prompt — with GPT-5.4 Mini. Only the number of questions per month changes between rows. Use the calculator above to model your own usage or a different model.

Usage levelQuestions / monthLowLikelyHigh
Light usage 100 $0.1290 $0.2025 $0.3000
Typical usage 500 $0.6450 $1.01 $1.50
Heavy usage 2,000 $2.58 $4.05 $6.00

Caveats

  • These estimates use the same reference token ranges as the site's chatbot calculator for question and answer length, accurate to ±10–15% for plain English text.
  • Each provider uses a different tokeniser. Actual token counts may differ ±5–10% between OpenAI, Anthropic, Google, Mistral, and DeepSeek for the same content.
  • Retrieval infrastructure cost (embeddings, vector database, reranking) is not included. This tool only estimates the cost of the model's answer-generation call itself.
  • Prices shown are manually verified against each provider's official pricing page as of 2026-09-08. AI provider pricing changes frequently — verify at your provider's official pricing page before making budget decisions.
  • "Focused"/"standard"/"broad" retrieval are this project's own descriptive labels for typical retrieved-context sizes, not a claim of any industry-standard terminology — use "Custom" and enter an exact token count for the most accurate estimate.
  • The context-window warning above appears when the High scenario's combined tokens exceed 80% of the selected model's context window — a safety margin before truncation or failure risk, not a hard limit at 100%.

This is an estimate, not a billing guarantee. Always confirm against your provider's usage dashboard and official pricing before committing a budget.