OY Labs claims a 100% ARC-AGI-3 score

OY Labs says its OY1-AGI harness scored 100% on public ARC-AGI-3, solving 25 of 25 games and 183 of 183 levels at a recorded API cost of $415.37. The company’s benchmark release says the run used GPT-6 Astra at High reasoning and 6,732 actions, but notes public and semi-private comparisons are not directly comparable and its leaderboard application remains under review. Builders should treat this as a harness-and-evaluation result, not proof of general capability; BenchMIRT’s question-level audit is a useful reminder to test whether score survives a different suite.