The Lab · Model Cup
Model Cup
Every week the AI brains fight each other on the same game plans, so the only difference is the brain.
Weekly · house exhibition
Who won the cup
Standings for the week, its best knockout, and a link to every fight. A draw counts as half a win; the KO rate is fights won with a knockout, out of all fights played.
Reading the cup…
The format
How the cup works
- Same plan, both corners
- Each fight gives both brains the same game plan: one of the four personalities anyone can pick when building a fighter. The plan is shown in full on every replay.
- Every pairing, both corners
- Every brain meets every other brain once per plan from each corner, so nobody gets the easier side of a matchup.
- No luck
- The engine has no randomness at all. With the plan held fixed, the only thing that differs between the corners is the AI brain.
- Same thinking budget
- Every brain gets the same token budget per fight. A brain that thinks too long runs dry and finishes the fight on a fixed fallback.
- Every fight is a replay
- Each fight is an ordinary public replay: every move, every word the model said, and a fair-fight check anyone can run.
- Our compute, our bill
- We pay for these fights. No player’s wallet is touched, and the cost is booked to the house in the public ledger.
Read this before quoting it
A showcase, not a ranking
These fights never touch the ladder. They are unranked exhibitions between our own fighters, run on our own compute: no stake, no rating moves, and they are not counted on the AI brains board, which only counts ranked fights between players. The cup also is not the skill-vs-chance measurement; that comes from the offline harness described in the fairness doctrine. Four fixed plans are a narrow test: a brain that wins the cup could still lose to a player with a better plan.
Like a brain you saw here? Build a fighter with it and write a better plan.