AI Cost Calculators
RAG Cost Calculator
Estimate the complete cost of a retrieval-augmented generation system across chunking, embeddings, vector storage, retrieval, reranking, LLM usage, refreshes, and setup.
Enter Your RAG Workload
Model ingestion, storage, retrieval, reranking, LLM usage, and setup in one place.
Knowledge Base and Chunking
Monthly Retrieval and LLM Usage
Current Provider Prices
Estimated RAG workload
Chunks per document: 3
Total vectors: 30,000
Initial embedding tokens: 13,500,000
Estimated vector storage: 0.26 GB
Refreshed vectors monthly: 3,000
Monthly queries: 50,000
Estimated RAG Cost
The result separates one-time setup from continuing monthly operating cost.
Monthly planning cost
Initial setup
—
Operating cost per query
—
Per 1,000 queries
—
Monthly refresh embeddings
1,350,000 embedding tokens
Rate not entered
—
Monthly vector writes
3,000 refreshed vectors
Rate not entered
—
Vector storage
0.26 GB estimated storage
Rate not entered
—
Vector reads
50,000 billable read units
Rate not entered
—
Reranking
50,000 monthly searches
Rate not entered
—
LLM input
107,500,000 input tokens
Rate not entered
—
LLM output
25,000,000 output tokens
Rate not entered
—
Fixed monthly costs
Subscriptions, hosting, monitoring, or platform fees
Rate not entered
—
Amortized setup
$0.00 spread across 12 months
Rate not entered
—
Monthly operating cost: —
Setup added per month: —
First-year total: —
Price inputs entered: 0 of 9
Budget status: Add a budget to compare
* Important: This calculator stores no built-in provider price. Enter the current official rates for your embedding model, vector database, reranker, LLM, and other services. Blank price fields are treated as zero. Storage is an estimate because compression, replicas, indexes, and provider billing rules can differ.
A RAG system has more cost layers than the final LLM response. Documents must be chunked and embedded, vectors must be stored and queried, content may need reranking, and the retrieved context increases the tokens sent to the language model.
Planning the Full RAG Cost Before Building
Start with the size of the knowledge base, average document length, chunk size, overlap, embedding dimensions, and storage format. The calculator estimates the number of chunks, initial embedding tokens, vector count, and storage required.
Then add monthly queries, retrieval depth, context tokens, LLM output, content refreshes, and current provider prices. The result separates initial setup cost from the continuing monthly operating cost.
This makes it easier to compare a managed RAG stack with a custom combination of embedding, vector database, reranking, and LLM providers.
How Chunking Changes Cost
Smaller chunks can increase the number of vectors in the database. Larger overlap repeats more source text between chunks. Both choices can increase initial embedding cost, storage, write operations, and monthly refresh cost.
Retrieval depth also matters. A larger top-k value can send more context into the language model and may increase reranking work. The best setting is not always the largest one; it should balance answer quality, latency, and cost.
Using Current Provider Prices
All monetary fields are blank by design. Enter the current official rate for the exact embedding model, vector database, reranker, and language model you plan to use.
Vector databases may charge by storage, reads, writes, capacity, serverless units, replicas, or a fixed plan. Convert those charges into the matching fields and place any remaining subscription amount under fixed monthly costs.
Blank price fields are treated as zero, so the calculator can also model a stack that does not use every cost layer.
Practical Decisions This Tool Supports
- Estimate RAG setup cost before development.
- Calculate vector count and approximate storage.
- Compare different chunk sizes and overlap settings.
- See how top-k retrieval changes LLM input cost.
- Plan monthly document refresh and reindexing cost.
- Calculate operating cost per query and per 1,000 queries.
- Compare a managed platform with a custom RAG stack.
- Check whether the planned system fits a monthly budget.
Understanding the Result
Initial setup includes the first embedding pass, initial vector writes, and any other one-time implementation cost entered. Monthly operating cost includes refresh embeddings, refresh writes, vector storage, reads, reranking, LLM usage, and fixed monthly fees.
Monthly planning cost adds an amortized share of the initial setup cost. First-year total keeps setup separate and adds twelve months of operating cost.
Frequently Asked Questions
What costs should a RAG estimate include?
A complete estimate can include initial embeddings, vector writes, storage, monthly re-embedding, database reads, reranking, LLM input and output tokens, fixed platform fees, and one-time setup work.
Why does chunk overlap increase cost?
Overlap repeats part of the source text across neighbouring chunks. This can improve continuity, but it also increases embedding tokens, vector count, storage, and refresh work.
How is vector storage estimated?
The calculator uses vector count, vector dimensions, bytes per dimension, metadata bytes, and an index-overhead percentage. The result is an estimate because providers may compress or replicate data differently.
What should I enter for vector read units per query?
Use the billable read or query units defined by your vector database. If the provider charges one query operation for each request, enter 1. If billing depends on scanned data or capacity units, convert your expected workload into the provider's billable units.
Why are the price fields blank?
RAG stacks can combine many providers, plans, regions, and private quotes. Blank fields prevent example prices from looking like verified live prices. Enter the current official rates for the exact services you plan to use.
Does the cost per query include setup cost?
The displayed operating cost per query excludes one-time setup. The monthly planning cost separately adds the setup amount spread across the selected amortization period.