By J. Wren · Last reviewed 2026-10-13
LLM Calc is a set of free, fast, browser-based calculators for people building with AI — LLM API cost estimators, token counters, context-window checks, and write-ups on how LLM pricing actually behaves.
Every tool runs entirely in your browser: no signup, no upload, nothing stored on a server. The goal is simple utilities that load instantly and just work.
Model pricing is re-read once a day from a public, community-maintained dataset rather than typed in by hand, so the numbers stay close to current. Each pricing page names its source and links to the vendor's official page — always confirm there before budgeting.
New tools are added regularly. If there's something you'd find useful, say so on the contact page.
A pricing site is only worth reading if someone is still watching it, so the maintenance is deliberately boring and automated. A scheduled job re-reads the upstream dataset once a day, rebuilds the affected pages, and redeploys. Runs are occasionally delayed, so the sync date printed on each page is the thing to go by. If the job fails it opens an issue rather than failing silently. An earlier version failed unnoticed for five days, which is exactly the failure this site exists to avoid, and the alerting was added because of it.
Two checks sit between the upstream data and what you see. Implausible movements are refused automatically: a rate that jumps by more than half, or a context window that moves by more than a fifth, is held back and verified against the vendor's own page by hand before it is allowed to publish. And every page is structurally validated on each deploy, after a bulk edit once stripped the head section from six pages and shipped the damage. Changes that pass are recorded with dates in the pricing change log, so the history is auditable rather than something the site quietly rewrites.
Guides that report measurements publish the script and the results file next to the article, so the figures can be re-run instead of taken on trust. The methodology page sets out where the data comes from, what the job checks, how "Last reviewed" is defined and what the site leaves out.
LLM Calc is written and maintained by J. Wren, a developer who builds projects on top of LLM APIs and kept losing track of what the calls actually cost, comparing vendor pricing pages by hand more often than felt reasonable. The site exists to do that arithmetic once, against rates that are re-read once a day instead of a page someone typed in and forgot about. J. Wren is one person, not a company. This is an independent site: the numbers are checked by scripts anyone can re-run rather than by credentials.
Most LLM pricing calculators on the web are typed in by hand and then abandoned. Rates change, models are replaced, and the page quietly keeps showing figures that were right a year ago — which is worse than showing nothing, because the numbers still look authoritative. The pricing here is generated from a data file that a scheduled job refreshes, so it goes stale only if the job stops, and the sync date is printed on the page so you can tell.
Every page is static HTML with a small amount of JavaScript, served from a CDN. There is no backend, no account system and no database. Calculations happen in the browser, which is why the pages load instantly and why nothing you type reaches a server.
Model rates come from a public, community-maintained dataset rather than from scraping vendor pages, because the vendors publish no pricing API and their pricing pages are rendered in ways that break automated reading. That choice is a trade-off, and it is stated on every page that uses the data, along with a link to the official source for verification. If a figure disagrees with the vendor, the vendor is right — and a short email gets it corrected.
They estimate list-price cost from token counts you supply. They do not know your actual usage, and they do not model prompt caching, batch endpoints or committed-use discounts, all of which can put a real invoice well under list price. Token counts are estimates too: the common four-characters-per-token rule is reasonable for English prose and poor for code, JSON and non-Latin scripts. Nothing here is a substitute for the vendor's own billing dashboard, and no figure on this site should be the last number you check before signing off a budget.
People building with LLM APIs who need a number before they commit: what a feature will cost at volume, whether a prompt fits, which rate limit binds first.
This site is run by one person, not a company. The domain is paid for out of pocket. Ads may be added later to cover that cost; they will not change which models are listed or what the numbers say. No vendor pays for placement, and the model list is chosen for coverage rather than for any commercial relationship — there are none. What the site collects, and what it doesn't, is set out in the privacy policy.