Abstract Visual Reasoning on ARC-AGI 2
100Accuracy (Pass@2)best.human
Evaluation results
| Method | Links | ||
|---|---|---|---|
| best.human2026.02 | 100 | — | |
| AEVO ProcedureCategory=Ours, Model=Gemini-3-Flash2026.05 | 47 | 12 | |
| ADASCategory=Procedure-Based Evolution, Model=Gemini-3-Flash2026.05 | 36 | 3 | |
| AFlowCategory=Procedure-Based Evolution, Model=Gemini-3-Flash2026.05 | 31.8 | 14 | |
| DGMCategory=Procedure-Based Evolution, Model=Gemini-3-Flash2026.05 | 29.8 | 5 | |
| Bespoke (Grok-4)#Params=1.7T2026.02 | 29.4 | — | |
| SPOCategory=Procedure-Based Evolution, Model=Gemini-3-Flash2026.05 | 25 | 6 | |
| GEPACategory=Procedure-Based Evolution, Model=Gemini-3-Flash2026.05 | 22.5 | 13 | |
| ReAct (Pass@1)Category=Single-Agent Inference, Model=Gemini-3-Flash2026.05 | 21.8 | — | |
| Grok-4-thinking#Params=1.7T2026.02 | 16 | — | |
| Loop-ViT (Large)#Params=18M2026.02 | 14.2 | — | |
| Loop-ViT (Medium)#Params=11.2M2026.02 | 11.5 | — | |
| VARC (ensemble)#Params=73M2026.02 | 11.1 | — | |
| Loop-ViT (Small)#Params=3.8M2026.02 | 10 | — | |
| VARC#Params=18M2026.02 | 8.3 | — | |
| TRM#Params=7M2026.02 | 7.8 | — | |
| HRM#Params=27M2026.02 | 5 | — | |
| o3-mini-high2026.02 | 3 | — | |
| GPT-52026.02 | 1.9 | — | |
| Deepseek R1#Params=671B2026.02 | 1.3 | — | |
| Claude 3.7 8k2026.02 | 0.9 | — | |
| Bespoke (Grok-4)#params=1.7T, type=large language models (LLMs)2026.04 | 0.294 | — | |
| Grok-4-thinking#params=1.7T, type=large language models (LLMs)2026.04 | 0.16 | — | |
| Enc-Dec ViT+KoPE*#params=37M, continual pretraining on ARC-AGI-2 training tasks=true2026.04 | 0.102 | — | |
| Enc-Dec ViT+KoPE#params=37M2026.04 | 0.098 | — | |
| ViT#params=18M, type=vision models (VARC)2026.04 | 0.083 | — | |
| TRM#params=7M, type=recurrent models2026.04 | 0.078 | — | |
| Enc-Dec ViT#params=36M2026.04 | 0.073 | — | |
| HRM#params=27M, type=recurrent models2026.04 | 0.05 | — | |
| o3-mini-high#params=N/A, type=large language models (LLMs)2026.04 | 0.03 | — | |
| GPT-5#params=N/A, type=large language models (LLMs)2026.04 | 0.019 | — | |
| Deepseek R1#params=671B, type=large language models (LLMs)2026.04 | 0.013 | — | |
| Claude 3.7 8k#params=N/A, type=large language models (LLMs)2026.04 | 0.009 | — |