Pricing tool

AI token cost calculator

What an LLM feature costs to run — per answer, per month and per year — with the retrieved context on its own line, because that is where most of a RAG bill hides. Enter your model’s prices and the shape of one request. Nothing you type leaves the page.

Estimate your running cost

The two prices are example values, not a price list — replace them with your model’s current price per million tokens. Everything is USD.

Your model’s prices, USD per 1M tokens

Input tokensThe model’s list price for input; example value

Output tokensUsually several times the input price; example value

One model call

Prompt and question, tokensSystem prompt, instructions and the user’s question

Retrieved passagesChunks your retrieval adds to the prompt; 0 without RAG

Tokens per passageYour chunk size

Answer length, tokensWhat the model writes back

Each request, and how many

Model calls per request1 for a RAG answer; one per step for an agent

Retrieval cost per request, USDEmbedding and vector search; $0.002 is what ours measures

Requests per month40,000 is the volume of the system we measured

How the cost is worked out

Four lines of arithmetic, all on the page, so you can check them rather than trust them.

Input tokens per call are the prompt and question plus the retrieved passages: prompt + passages × tokens per passage. They are priced at the model’s input rate, and the context share is shown apart from the prompt.

Output tokens per call are the answer, priced at the output rate.

Per request, both are multiplied by the model calls the request makes — one for a RAG answer, one per step for an agent — and the retrieval cost is added once.

Per month is the per-request cost times requests a month; per year is twelve months of that. Rounding happens only for display.

What one of ours costs to run

A measured system, so you have a real number to set yours against.

The production platform behind our enterprise AI case study averages about $0.011 per answer across roughly 40,000 queries a month: about $0.002 of retrieval and $0.009 of generation. Multiplied out, that is about $440 a month — arithmetic on the two measured figures, not a third measurement. Answers take 1.9 seconds on average and 4.2 at p95, to the last token.

The split is the part worth carrying to your own case. The retrieval infrastructure is the cheap fifth. What retrieval really costs sits inside the generation line, as the input tokens every retrieved passage adds to every call. Context length is the cost lever; a cheaper vector database is not.

How that RAG system was added to a product · What an agent costs per task

Three ways to cut the bill

In the order they usually pay off.

Shorten the context

Fewer, better-ranked passages cut cost on every call. Try halving the passages above and watch the context line — it is usually the biggest one you control.

Make fewer calls

If an agent’s steps are always the same, it is a pipeline with one model call in it: cheaper, faster and testable line by line.

Route the easy work

Classification and routing steps rarely need the largest model, and repeated lookups can be cached instead of asked again.

Common questions

About the running cost of an LLM feature.

How do you calculate the cost of an LLM answer?

Count the input tokens of one model call — the prompt, the question and any retrieved passages — and multiply by the model’s input price per million tokens; count the output tokens and multiply by the output price. Multiply by the number of model calls one request makes, add whatever retrieval costs per request, and multiply by requests a month.

Why does RAG make LLM answers more expensive?

Because every retrieved passage arrives as input tokens on every call. On the production system we measured, an answer costs about $0.011, and $0.009 of it is generation — most of that the retrieved context — while the retrieval infrastructure itself is about $0.002. Fewer, better-ranked passages cut the bill more than a cheaper vector database does.

How many tokens is a page of text?

It depends on the language and the layout, so measure your own. For scale, the corpus behind our enterprise AI case study — about 15,000 pages in five languages — came to 10–15 million tokens, which is roughly 700 to 1,000 tokens a page.

What does an AI agent cost per task?

Every step an agent takes is another model call, so set model calls per request to the number of steps. The cost and the waiting time grow with the steps, which is why the cheapest agent is the one with the fewest steps that still does the job. What AI agent development costs.

Does this include the cost of building the feature?

No. This is the running cost, which is a third-party cost passed through at cost. Building an LLM feature into a product you already run is $15,000–$30,000 over 2–4 weeks; a full AI product is $40,000–$120,000 over 8–12 weeks. The AI development cost calculator places a project inside those bands.

Want this number for your product?

Tell us what the feature does and how many people will use it. We come back with the build price and the running cost, in writing, within 48 hours of the call.

Name *

Work email *

What you’re building *

I agree to the Privacy Policy and consent to Yarvixo processing my data in accordance with GDPR.

Message received

We’ll review your enquiry and respond within one business day.