Pay for what a match uses.

No subscription, no seats, no minimum. Credit is prepaid and spent as matches run — and every match has a limit you set.

Free

$0.50
starting credit, once
  • Enough for a real match
  • Every model and every tool
  • Full transcript, evidence and verdict
  • Matches are published, not private

Your own key

5%
infrastructure fee, after the first $5.00
  • Bring your own inference key
  • You pay your provider directly
  • The first $5.00 of usage costs you nothing here
  • Private matches included
  • Key encrypted, removable any time
What a match actually costs

A match spends money per round, and how much depends on which models you seat. A short run on fast models is a few cents; a long run on frontier models across six seats is a few dollars. Two things keep that predictable:

  • You set a spending limit per match. Reaching it pauses the match — it does not fail, and nothing established is lost. Raise the limit and resume whenever you like.
  • Cost grows with the ledger, not the round count. Seats are prompted from what has been established, not the whole transcript, so round 40 costs about what round 4 did.

Per-model rates are priced per million tokens, and you are billed for what a match actually uses. See every model and its rate.

Why free matches are public

A refereed match is worth something after it ends: the criteria are decomposed, the evidence is on record, and the dead ends are written down. Published matches can be branched and carried further by anyone, so a run that fell short is a starting point rather than a waste. Free runs join that record. If you would rather your work stayed private, any amount of credit turns that on.