◢◤

deus ex machina

Where does the same open model cost 13× more depending on who serves it?

loading…

Cache hit rate 0%

Hide single-provider models

Providers vs. average price

Each dot is a model · x = number of paid providers · y = average price (log scale) · dot size = token consumption · dashed line = least-squares trend (log y).

#	Model	Providers	Tokens (30d day)	Cheapest	Priciest	Spread	Price range ($/1M)	Variance (CV)

Prices in USD per 1M tokens. "Blended" = (3×input + 1×output) / 4, where input is cache-aware: effective input = fresh×(1−hit) + cached×hit. Providers that don't publish a cached-read price get no discount (cached tokens billed as fresh), so raising the cache hit rate can change which provider is cheapest and widen or narrow the spread. Spread = priciest ÷ cheapest paid provider. Variance shown as the coefficient of variation. Free ($0) endpoints are excluded from spread/variance math. Cache write (a one-off cost to store a prefix, charged by only a few providers) is shown per provider but not blended in, since its impact depends on prefix-reuse count.

Source: OpenRouter (per-provider pricing + token-consumption rankings). Artificial Analysis benchmarks shown when an API key is configured. Data refreshes daily via GitHub Actions.