AI inference spend is rising faster than any team can track. There's no benchmark to check an invoice against, no reference rate to negotiate a contract from, and no way to forecast next quarter's bill. Engineers can't tell you if the model is worth the price. Finance can't tell you if the price is fair.
See how ACPI worksEvery model tracked by ACPI, in one table — live price per million tokens, provider, blended rate, and context window, refreshed hourly from 60 sources.
No bundled SKUs, no vendor-quoted rate cards. Filter, sort, and compare the same numbers your team uses to decide what to actually run.
Explore the full screenerRoute through the gateway and every request is benchmarked against ACPI in real time — no blended contract to renegotiate, no dashboard that only shows the damage after the invoice lands.
One line of code. No per-model SKUs, no annual commit to unlock visibility, no waiting on finance to reconcile the bill.
Connect your stackACPI blends live pricing across every tracked model, weighted by benchmark-derived quality and split between frontier and commodity tiers — not a flat average of list prices.
Live rates pulled hourly from 60 providers, public and metered.
Each price is weighted against benchmark performance, not averaged flat.
361 models collapse into one reference number — the market rate.