Action-efficiency Game Solving on ARC-AGI 3 (public evaluation)
78.4Score (%)OPINE-World
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| OPINE-WorldNote=no per-game training, Base Model=Claude Opus 4.82026.07 | 78.4 | 20 | 90 | |
| baseline1Note=single-agent coding agent, Base Model=GPT-5.52026.07 | 63.8 | 14 | 80 | |
| Vision†Note=pre-trains on eval set, Protocol=continual-learning2026.07 | 63.2 | 12 | 73 | |
| WorldCoderNote=clears no level2026.07 | 0 | 0 | 0 | |
| Latent world modelsNote=clears no level2026.07 | 0 | 0 | 0 |