Loading…
Estimate your monthly retrieval-augmented generation costs. Adjust the inputs to match your corpus size and query volume. These are rough estimates. Actual costs depend on your chosen providers and architecture.
Total chunks
2,000
~Tokens / chunk
188
Context per query
~2,000 tok
LLM model
GPT-5
| Line item | Cost |
|---|---|
| One-time embedding text-embedding-3-small · paid once, re-run on updates | < $0.01 |
| Monthly re-embedding 10% of corpus refreshed | < $0.01 |
| Vector storage 2,000 vectors on managed service | $0.16 |
| LLM input tokens 5,000 queries × ~2,000 tok context | $12.5 |
| LLM output tokens 5,000 queries × ~400 tok response | $20.0 |
| Total monthly excluding one-time embedding | $32.7 |
These are estimates.
Actual costs vary depending on your vector database provider, LLM choice, reranker usage, and query patterns. A production RAG system at moderate scale typically runs $100–$800/month in infrastructure. Book a call to get a real cost estimate for your specific architecture.
Get a real quoteYou pay to embed your document corpus once using a model like text-embedding-3-small ($0.02/1M tokens). You re-embed when documents change. A 1,000-page corpus (~750k words) costs roughly $0.50 to embed in full.
Pinecone, Qdrant, Weaviate, and pgvector all charge per stored vector or per index size. A 100k-chunk corpus costs $25–$70/month on managed services. Self-hosted is cheaper but requires ops overhead.
Each user query passes retrieved chunks (context) to an LLM. With GPT-5 at $1.25/1M input tokens, a query with 2,000 tokens of context costs ~$0.0025. At 10,000 queries/month that's $25.
Reranking model costs, chunking/preprocessing compute, monitoring infrastructure, or engineering time. A production RAG system typically runs $100–$800/month in infrastructure at moderate scale.