About LLM Cost Check

An independent, browser-based token counter and cost calculator for large language model APIs — built because every existing option answered a narrower question than the one people actually have.

What this site is for

Anyone building on a large language model API eventually hits the same problem: the pricing pages are easy to read and impossible to apply. A provider will tell you a model costs $3.00 per million input tokens. What you want to know is whether the thing you are about to build costs $40 a month or $4,000, and whether a cheaper model would do the same job.

Getting from one to the other requires several steps that are individually simple and collectively annoying: work out how many tokens your prompt is, remember that each provider tokenizes differently, estimate how long the responses will be, remember that output is billed at several times the input rate, multiply by your expected volume, and then — if you are building anything conversational — account for the fact that chat APIs resend the entire conversation history on every single turn.

The calculator does all of that in one screen. You paste the prompt you actually intend to send, set your expected response length and monthly volume, and get a per-run and per-month figure for thirteen models side by side, along with a check on whether your prompt even fits each model's context window.

The guides section exists for the questions that follow. The calculator can tell you that output tokens cost more; it cannot explain the prefill/decode asymmetry that causes it, or walk you through structuring a prompt so that prompt caching actually hits. Those are long-form answers, so they live in long-form articles.

Who builds it

LLM Cost Check is a small independent project, built and maintained by one developer rather than a company. It is not affiliated with OpenAI, Anthropic, Google, Meta, xAI, DeepSeek, or any of the gateway platforms listed on the site, and none of them have any input into what appears here.

That independence is the point. A cost comparison published by a model provider will always have a thumb on the scale. This one has no reason to prefer any particular answer, which is why the calculator will happily tell you that the cheapest model that fits your prompt is one of the $0.14-per-million options rather than a flagship.

How the numbers are produced

The short version:

OpenAI models are tokenized exactly, in your browser, using OpenAI's real o200k_base encoding via js-tiktoken. These results are labelled Exact.
Every other model is estimated from a characters-per-token ratio tuned per provider, because no provider outside OpenAI publishes a tokenizer that can run client-side without a network call. These results are labelled ~Approx and typically land within a few percent.
Pricing is transcribed by hand from each provider's published rate card and reviewed regularly. The date of the last review is shown in the calculator's footer.

The long version, including the specific ratios used, how multi-turn and cached costs are modelled, and an honest list of the places where these numbers are least reliable, is on the methodology page. If you are going to use this tool to make a real budgeting decision, read that page first.

What this site deliberately does not do

It does not send your text anywhere. Tokenization and cost arithmetic both run as JavaScript in your browser. There is no backend, no database, and no endpoint that receives prompt text — which means you can safely paste a real system prompt or a confidential document into it. This is a design constraint rather than a policy promise: there is no server to send it to.

It does not track you. No analytics, no accounts, no first-party cookies. The only third-party code on the site is Google AdSense, which is how the site is funded; what that involves is set out in the privacy policy.

It does not benchmark quality. The calculator answers "what would this cost on each model", not "which model is best". Those are different questions, and cost comparison is useless if you treat it as a substitute for evaluating whether a cheap model can actually do your task. Always test capability before optimising price.

It does not replace the provider's own rate card. Prices here are maintained manually and providers change them without notice. Treat the figures as a well-maintained estimate for comparison and planning, and confirm current rates at the source before signing off a production budget.

How it is funded

The site is free and shows advertising through Google AdSense. There are no affiliate links in the model or gateway tables — the links to provider pricing pages are plain links, and no platform pays to appear in the comparison or to be ranked more favourably within it. If that ever changes, it will be disclosed on this page.

Corrections

Pricing data goes stale, tokenizer ratios drift when providers ship new models, and a manually maintained table will eventually contain a mistake. If you spot one, please report it — corrections are the most useful contribution anyone can make to a site like this, and they get fixed quickly.