Reasoning on ARC Mini
97.32AccuracyQwen3 + DIARC
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen3 + DIARC#Params=3.6B2026.06 | 97.32 | |
| Qwen3 + SFT only#Params=3.6B2026.06 | 95.3 | |
| Minitron + DIARC#Params=6.9B2026.06 | 76.51 | |
| Minitron + SFT only#Params=6.9B2026.06 | 74.5 | |
| Llama-3.2 + DIARC#Params=2.6B2026.06 | 73.15 | |
| Llama-3.2 + SFT only#Params=2.6B2026.06 | 70.47 | |
| EqR2026.05 | 55.28 | |
| TRM2026.05 | 48.35 | |
| HRM2026.05 | 44.85 | |
| o1-mini#Params=N/A2026.06 | 30.2 | |
| DeepSeek-R1#Params=671B2026.06 | 28.86 | |
| GPT-4#Params=N/A2026.06 | 24.1 | |
| Qwen-2.5#Params=7B2026.06 | 22 | |
| Minitron#Params=8B2026.06 | 6.04 | |
| Llama-3.2#Params=3B2026.06 | 1.34 | |
| Qwen3#Params=4B2026.06 | 0.67 |