Science Question Answering on ARC Challenge (Accuracy and Token Cost)
70.1AccuracyHyPER
Evaluation results
| Method | Links | ||
|---|---|---|---|
| HyPERModel=OLMoE2026.02 | 70.1 | 0.68 | |
| DeepConfModel=OLMoE2026.02 | 68.7 | 0.32 | |
| Self-CertaintyModel=OLMoE2026.02 | 66.4 | 0.76 | |
| DeepConfModel=DeepSeek-V22026.02 | 66.2 | 0.53 | |
| HyPERModel=DeepSeek-V22026.02 | 66.2 | 0.74 | |
| RoEModel=OLMoE2026.02 | 65.2 | 1.67 | |
| SCModel=DeepSeek-V22026.02 | 65.2 | 1 | |
| Self-CertaintyModel=DeepSeek-V22026.02 | 64.7 | 1.43 | |
| RoEModel=DeepSeek-V22026.02 | 63.7 | 1.65 | |
| SCModel=OLMoE2026.02 | 63.3 | 1 |