Abstraction and Reasoning on ARC-AGI 2 (public evaluation)
100Pass@2Human Panel
Evaluation Results
| Method | Links | |
|---|---|---|
| Human PanelCategory=Human2026.04 | 100 | |
| CoreThink Meta-ClassifierCategory=Neuro-Symbolic + Ensemble2026.04 | 30.8 | |
| J. BermanCategory=Hybrid2026.04 | 29.4 | |
| NVARCCategory=Hybrid2026.04 | 27.6 | |
| Compositional ReasonerCategory=Neuro-Symbolic2026.04 | 24.4 | |
| GPT-5-ProCategory=LLM2026.04 | 18.3 | |
| Grok-4 (Thinking)Category=LLM2026.04 | 16 | |
| Claude Opus 4 (16K)Category=LLM2026.04 | 8.6 | |
| o3 (High)Category=LLM2026.04 | 6.5 | |
| o4-mini (High)Category=LLM2026.04 | 6.1 | |
| Claude Sonnet 4 (16K)Category=LLM2026.04 | 5.9 | |
| o3-Pro (High)Category=LLM2026.04 | 4.9 | |
| Gemini 2.5 Pro (32K)Category=LLM2026.04 | 4.9 |