Mathematical Reasoning on AIME 26 (val)
75.2Accuracy (%)Base
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| BaseBackbone=Qwen3-32B, Generations=642026.04 | 75.2 | 3.24 | 26.7 | |
| ControlBackbone=Qwen3-32B, Generations=642026.04 | 74.1 | 3.83 | 35.2 | |
| MEMENTOBackbone=Qwen3-32B, Generations=642026.04 | 72.6 | 1.67 | 14 | |
| BaseBackbone=Phi-4-r (14B), Generations=642026.04 | 71.7 | 2.65 | 28.8 | |
| ControlBackbone=Phi-4-r (14B), Generations=642026.04 | 69.8 | 3.04 | 28.6 | |
| BaseBackbone=Olmo 3 (7B), Generations=642026.04 | 67.9 | 3.95 | 50.8 | |
| MEMENTOBackbone=Phi-4-r (14B), Generations=642026.04 | 67.6 | 1.17 | 11.3 | |
| BaseBackbone=Qwen3-8B, Generations=642026.04 | 66.8 | 2.41 | 25.3 | |
| Mem. + RLBackbone=Qwen3-8B, Generations=642026.04 | 64.9 | 1.45 | 14.9 | |
| ControlBackbone=Qwen3-8B, Generations=642026.04 | 64.7 | 2.59 | 28.3 | |
| ControlBackbone=Olmo 3 (7B), Generations=642026.04 | 59.8 | 3.51 | 37.1 | |
| MEMENTOBackbone=Qwen3-8B, Generations=642026.04 | 57.3 | 1.02 | 9.7 | |
| MEMENTOBackbone=Olmo 3 (7B), Generations=642026.04 | 55.4 | 3.21 | 37.8 |