If the score holds on the private dataset, Schema demonstrates models have latent intelligence unlocked through better harness engineering.
Schema scores 98.98 and 95.35 percent on the ARC-AGI-3 public evaluation set
If the score holds on the private dataset, Schema demonstrates models have latent intelligence unlocked through better harness engineering.
Schema scores 98.98 and 95.35 percent on the ARC-AGI-3 public evaluation set
Published on July 17, 2026