How to Use Measured Lab Data

Measured Lab records AI API prices, subscription limits, tool plans and GPU rental offers, and runs its own API test requests. Every value carries its date, source and data status. Methodology · Coverage.

Who the data is for

• Developers. Compare official API prices and OpenRouter routes, then check measured speed for the country, model version and route you plan to use. Use the cost calculator to apply your own input and output token counts.
• Teams choosing a subscription. Read the published price, billing period, included usage and limits together. A web subscription does not necessarily include API usage. Local prices are shown only when read in that country; a missing local price is never replaced with a US price.
• Procurement teams. Compare dated records, the named price source and the history of confirmed changes. Provider claims about features are separate from our own API measurements.
• Journalists and researchers. Cite the observation date, model version, route, country and sample size. The measured sample is not the whole market. Use the linked JSON and CSV downloads to check a figure.

How to read statuses

• Observed — our API test requests, with their route, country, date and sample size.
• Published — a price or limit read from the linked provider page; a provider claim is labeled as such.
• Calculated — arithmetic using the stated prices and your inputs; assumptions are shown.

Data readiness is separate: Ready means the required history is available; Collecting means we are still gathering checks on two different dates; Not measured means no eligible observation is available. Unknown values stay unknown.

A fault on our side is recorded as a measurement error. Results that require two dates need successful measurements on two separate dates.

Read the scope before comparing

API prices are per million input or output tokens unless another unit is stated. Cache reads, cache writes, batch requests and long-context tiers can have different rates. Calculators show their inputs and assumptions; their result is an estimate, not a bill.

Speed measurements describe our own API requests. Time to first token, output tokens per second and total request time measure different things. Compare the same route, country and workload. The monthly quality check covers a fixed instruction-following sample; it is not a general ranking of model ability.

Cite and reuse the data

Use the page title, observation date and page URL, plus the sample size and route when citing a test. Our own measurements and calculations are licensed CC BY 4.0 with attribution to Measured Lab. Provider quotations belong to their owners. The data downloads page lists the available files and formats.

Corrections

Send the page address, disputed value, observation date and a supporting source to the contact on the corrections page. Read about the publisher and the methodology before using the figures in a decision.