
See https://arcprize.org/arc-agi-pub.
This allows for arbitrary API usage unlike the ARC-AGI prize.
I will resolve based on when the date when the method is first reported even if it is not submitted until substantially later (due to e.g. no public API).
I will apply a restriction roughly in line with the $10,000 runtime cost contraint, though it is unclear how this should resolve with respect to private models with no known API price.
I will exclude solutions that seem heavily overfit based on the semi-private evaluation set.
(Cut off data is resolved based on midnight PT.)
People are also trading
I think this market should resolved yes based on https://x.com/arcprize/status/1999182732845547795 and https://arcprize.org/leaderboard. Any objections?
@RyanGreenblatt I think this resolves NO? o3 did pass 85%, but it was well above the cost constraints.