Skip to content
inference.academy

Colophon

Where the numbers come from and what they do not claim.


Catalogue data
Model, endpoint, price, quantization and context data comes from the public OpenRouter API, swept nightly across every model it lists. Prices are vendor-reported, not measured.
The time series
Each sweep is archived, so the interesting question is not today’s price but which host quietly changed quantization, moved price, or dropped a model. That series only accrues with wall-clock time, which is why it started before the site did.
Cross-checks
Where a figure is compared against Artificial Analysis, it is cited as theirs. Nothing here republishes their dataset.
What is measured
Latency and throughput figures are reported by the router over a rolling window, not matched-condition measurements. Anything presented as measured was run through the compound harness under stated conditions, and says so.
Type
IBM Plex Sans and IBM Plex Mono. The wordmark is Plex Mono, outlined.
Built with
Next.js, Tailwind CSS, and no analytics.