AI Engineer Ready · Webinar Kit · Tool 1 of 3

LLM Cost Calculator

Cost an AI system before you build it, in rupees, the way it actually lands on an Indian card. Start from one of the six blueprints, then change chunking, caching and the model and watch what moves the bill.

1 · Start from a blueprint

Each preset loads realistic defaults for that project. Every number below stays editable.

2 · Model & traffic

Volume is the first lever. Most people guess it 10× wrong in either direction.

Model Requests per dayquestions, tickets, runs… Active days per month22 = weekdays only LLM calls per request1 = single call · 4 = four-agent pipeline Output tokens per call≈ 0.75 words per token

3 · What goes into every call

The prompt is never just the question. This is where the money goes.

System prompt + tool definitionsstable text, same on every call User input per callquestion, ticket body, log extract… Conversation history per call0 for one-shot tasks

4 · Documents & chunking

Turn this on for anything with RAG. Chunk size × top-k is the hidden multiplier on every request.

Retrieval (RAG) is part of this systemRetrieved chunks get pasted into the prompt

Chunking strategy

Chunk size (tokens) Top-k chunks sent to the model Number of documents Avg pages per document≈ 500 tokens per page Chunk overlap % Corpus re-embedded per month %docs that change and need re-indexing

5 · Prompt caching

The cheapest lever there is: the stable part of the prompt is billed at a fraction of the price when it is reused.

Cache the system promptClaude: cache reads ≈ 5–10% of input price, writes ≈ 1.25×

Cache hit rate %steady traffic ≈ 80–95%, bursty ≈ 40–60%

6 · How it lands in rupees

Your provider bills in dollars. What your card statement shows is higher.

USD → INR rate Card forex markup %most Indian cards charge 2–3.5%

Add 18% GSTOn foreign digital services billed in India

Estimated monthly bill

₹0

$0 at provider list price

Per request

₹0

Per 1,000 requests

₹0

One-time indexing

₹0

Where the money goes

Monthly cost by component, for the selected model.

Component	Tokens / request	₹ / month	Share

What just moved the number

Same system, every model

Identical inputs, monthly ₹. Click a row to switch model. checked = verified price · verify = confirm on the provider's page before you quote it.

Model	₹ / month	vs selected

Edit prices (USD per 1M tokens)

Prices change every few months. Update them here; the calculator recalculates instantly. Prices last checked: .

Model	Input	Cached read	Output

Embedding model price (USD / 1M tokens)

Not included: hosting, vector database, observability tools, retries after failures, and evaluation runs. In production, budget an extra 20–40% on top of the model bill for retries and evals alone.