Background

ARC-AGI was introduced in 2019 as a grid-based reasoning benchmark (“v1”) that tests whether AI can infer new rules from a handful of examples rather than rely on memorization; by the close of ARC Prize 2024, open-source solvers had plateaued around 53 % accuracy on the private set, while a high-compute run of OpenAI’s o3-preview model vaulted to roughly 75–88 %, underscoring v1’s saturation. In response, the ARC Prize Foundation unveiled the harder, human-validated “ARC-AGI-2” (“v2”) on 24 March 2025 and opened the 2025 Kaggle contest with a compute cap of about US $0.42 per task, yet kept the headline rule intact: the first fully open-source system to reach ≥ 85 % on the private v2 evaluation set wins the $1 Million Grand Prize. This prediction market will only track the year in which ARC publicly awards that Grand Prize.

Resolution Criteria

The market resolves to the calendar year in which the ARC Prize Foundation publicly announces and awards any portion of the Grand Prize to one or more teams.

• The Grand Prize is presently defined as the award unlocked when a submission reaches ≥ 85 % accuracy on ARC-AGI-2 during a formal annual competition period.

• If ARC introduces a new test or changes the required accuracy threshold, the criterion remains “the first year they actually pay out (or commit in a binding announcement to pay out) the prize labelled as the ARC Grand Prize.”