Home / Tools / RAG cost

RAG cost calculator

Estimate retrieval-augmented generation cost: corpus embedding, per-query retrieval pack, and LLM generation. See which line dominates your monthly bill.

OpenAI · input $0.400/1M · output $1.60/1M · cached $0.100/1M · rates checked 2026-08-12

Corpus indexing

Editable. Not from the LLM catalog. Default matches common small embed pricing.

0% = one-time index only. 10% amortizes a full re-embed every ~10 months into the monthly total.

Per query retrieval + answer

Retrieved context ≈ 2,560 tokens (chunk × k).

Applies published cached-input rates when the model supports them. Useful for a stable system prompt prefix.

Input tokens / query
3,040
Generation / query
$0.001856
Query embed / query
$0.0000016
Index once
$0.04
Generation / month
$55.68
Query embeds / month
$0.048
Index amortized / month
$0.004
Total / month
$55.732 (100% generation)
LinePer queryPer month
Generation (LLM)$0.001856$55.68
Query embedding$0.0000016$0.048
Corpus embed (amortized)n/a$0.004

How to use this

  1. Pick the generation model you will call.
  2. Set corpus size and your real embedding $/1M from the embed provider.
  3. Set chunk size and top-k so retrieved tokens match production.
  4. Enter system, question, answer length, and monthly queries. Read which line dominates.

Open GPT-4.1 Mini in cost calculator · RAG cost guide · Reduce RAG costs · Context window fit

Related tools

Embedding cost · Context window · Generation cost

Related guides

RAG cost calculator: estimate pipeline spend · RAG pipeline breakdown · Reduce RAG costs · Top-k and chunk cost · Embedding vs generation

Sources and references

Official documentation used for definitions, counting methods, or rate cards. Always confirm critical budgets on the provider page.

FAQ

What are the cost components of a RAG system?
Indexing (embed the corpus), tiny per-query embedding of the question, and generation (system + question + retrieved chunks + answer). Generation usually dominates.
Why is generation more expensive than embedding?
Embedding rates are often cents per million tokens and mostly one-time. Generation bills retrieved context as input on every query at LLM rates.
How do I reduce RAG costs?
Use a cheaper generation model when quality allows, lower top-k or chunk size, cache a stable system prompt, and cap answer length.
Are embedding rates from the TokenCalculator catalog?
No. Edit the embedding $/1M field yourself from your embed provider. Generation uses published catalog rates for the selected LLM.