About

LLM Cost Lab tracks what large language model APIs actually cost, and turns those rates into monthly numbers for real workloads.

Provider rate cards are published as a wall of per-million-token prices. That format hides the two things that decide your bill: the ratio of input to output tokens in your traffic, and how much of your input repeats often enough to be cached. A model priced at $2 per million and one priced at $10 per million are not five times apart on an agent workload with a 75% cache hit rate — they can be closer, or further apart, depending entirely on shape.

So this site does two things. It keeps a verified, dated price table across every major provider, and it runs that table through eight workload profiles that cover most production LLM spend. Every number on every page is computed from the same dataset, so the site is internally consistent by construction.

Independence

We are not affiliated with OpenAI, Anthropic, Google, or any other provider, and no provider pays for placement, ranking, or inclusion. Rankings are produced by arithmetic on published prices — there is no editorial thumb on the scale, because there is no editorial step.

The site carries display advertising, and links to third-party developer tooling in a clearly labelled section below the pricing tables. Some of those are affiliate links and are marked as such. Neither advertising nor affiliate relationships influence which model appears cheapest: that is decided by the cost formula on the methodology page, which you can check against the providers' own rate cards. No paid link ever appears inside a ranking or a price table — if one did, every other number here would be worth less.

Accuracy

Prices change without notice and this site can lag. Every page carries the date its prices were last verified. Treat the figures as planning estimates, confirm anything you are about to commit real money to, and tell us when something is wrong.