Cost-per-correct-answer benchmarking

Find the cheapest model that actually passes your tests.

Run your own cases across every model, score the answers, and rank by cost per correct answer — so a cheap model that passes beats an expensive one. The ranking engine is open-source; you only pay for the hosted runs you make.

Sign in · 5 free credits  ·  How it works  ·  Explore benchmarks  ·  Open-source engine on GitHub

Open-source engine

The open engine — local CLI/UI, model catalog, estimates, scoring, and cost-per-correct ranking logic — is yours to inspect or run yourself.

Hosted TokenHunger

This site runs that same engine with managed provider keys, GitHub sign-in, credits, saved analyses, and MCP access for agents.