Abstract Reasoning on ARC-AGI 2 (Semi-private)
100AccuracyHuman Panel
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Human PanelAuthor=Human, Comment=At least two humans out of ~400 solved it2026.06 | 100 | 17 | |
| Modality-Driven Search + Holistic Trace JudgingAuthor=Johan Land, Comment=Semi-private (official)2026.06 | 72.9 | 38.99 | |
| GPT-5.2 Pro (High)Author=OpenAI2026.06 | 54.2 | 15.72 | |
| Gemini 3 Pro (Refine.)Author=Poetiq2026.06 | 54 | 30.57 | |
| GPT-5.2 (X-High)Author=OpenAI2026.06 | 52.9 | 1.9 | |
| Gemini 3 Deep Think (Preview)Author=Google2026.06 | 45.1 | 77.16 | |
| GPT-5.2 (High)Author=OpenAI2026.06 | 43.3 | 1.39 | |
| GPT-5.2 Pro (Medium)Author=OpenAI2026.06 | 38.5 | 8.99 | |
| Opus 4.5 (Thinking, 64K)Author=Anthropic2026.06 | 37.6 | 2.4 | |
| Gemini 3 Flash Preview (High)Author=Google2026.06 | 33.6 | 0.23 |