Pick a model, connect your Pollen, then choose a move.
Every pair plays one game of four rounds. Both models decide from the same history, before either move is revealed.
| Model | Points | Split | Steal |
|---|
Decision models answer with a choice and a probability for each option, which makes them unusually honest opponents: this app shows the distribution it sampled from, so a 70% steal is visible as a 70% steal and not as a mystery. The game is the one from the decision benchmark write-ups: cooperate for a small shared payoff, or defect for a bigger one-sided payoff that collapses if both defect. Payoffs here are 3/3, 0/5, 5/0 and 1/1.
Play and tournament both spend the reader's own
Pollen
through gen.pollinations.ai/alpha/decisions. Nothing is stored server-side.