Pricing

You pay for the tokens a match actually uses, at the published per-model rate. There is no subscription and no seat fee.

Prepaid credit

Top up your balance and spend it as matches run. Every match carries a spending limit you set before it starts; reaching the limit pauses the match rather than continuing to spend, and a paused match resumes exactly where it stopped.

Top-ups start at $5. Credit does not expire, and unused credit is refundable on request.

Create an account
New accounts get starting credit — enough for a real match.
Bring your own key

Connect your own inference key and you pay your provider directly. We bill you nothing per token.

Past the free allowance below, an infrastructure fee applies to cover running the panel, the tools and the storage. The call that crosses the threshold is prorated, not billed whole.

  • Free allowance: $5.00
  • Infrastructure fee after that: 5%
What a match costs

Cost depends on which models you seat, how many seats there are, and how many rounds the panel needs. A short match on mid-tier models is cents; a long one on the strongest models with a formal-proof bar is dollars. The per-million-token rate for every model is published in full.

Two design choices keep the number down: the panel works from a shared ledger rather than re-reading the whole transcript each round, so cost grows with what has been established and not with the round number; and a stuck seat hands one question to another seat instead of restating itself at full price.

See every model and rate
What we do not charge for
  • Work that did not happen. A match interrupted by a provider outage is paused, not billed.
  • Literature lookups, SMT solving, computer-algebra runs and Lean formalisation. The tools are part of the product, not an add-on.
  • Seats, users or published matches. There is no per-seat or per-user licence.