Cost calculator
RAG Cost Calculator
Price the full retrieval-augmented generation loop: one-time corpus embedding, monthly re-embedding, query embeddings, retrieved chunks sent to the model, and the answer. See which knob, top-k, chunk size, or model, actually moves cost.
Estimated cost
Calculating…
Results update as you change the inputs.
Methodology and assumptions
Ingest = corpus tokens × embedding rate, paid once and shown separately. Monthly cost = re-embedding (corpus × churn) + query embeddings (queries × question tokens) + generation. Each generation prompt is system prompt + question + top-k × chunk tokens; the system prompt can be cached (hits at the cached-input rate, misses at the cache-write rate), while retrieved chunks change per query and are billed as uncached input. Long-context rates apply when the prompt exceeds a model's threshold.
Embedding prices are standard (non-batch) text rates. Vector database, reranking, hybrid search, retrieval compute, grounding/search tools, storage, networking, and taxes are excluded; add them for your stack. Formula version 2026-10-01.
Official sources:
Pricing last verified:
Official source: Vendor API pricing pages (all listed below)
Frequently Asked Questions
What does the RAG Cost Calculator calculate?
Price the full retrieval-augmented generation loop: one-time corpus embedding, monthly re-embedding, query embeddings, retrieved chunks sent to the model, and the answer. See which knob, top-k, chunk size, or model, actually moves cost.
How accurate is the RAG Cost Calculator?
It is a planning estimate based on the inputs, formula, assumptions, and sources shown on this page. Actual prices, invoices, taxes, discounts, and outcomes can differ.
Which pricing source does the RAG Cost Calculator use?
The calculator links to Vendor API pricing pages (all listed below) as its official source. The displayed pricing or reference data was last verified on 2026-10-08.
Are calculator inputs saved or sent to analytics?
Inputs are processed in your browser. Calculator values are not included in analytics events.