Scientific Problem Solving on FrontierScience-Olympiad
5.59Token Efficiency Ratio (B_method/BMV)ESC sweep
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| ESC sweepModel=GPT-OSS-120B, Target accuracy (alpha)=90%2026.05 | 5.59 | 33.8 | 50 | 107 | |
| PC-linearModel=GPT-OSS-120B, Target accuracy (alpha)=99%2026.05 | 3.88 | 33.8 | 50 | 2.5 | |
| ESC sweepModel=GPT-OSS-120B, Target accuracy (alpha)=75%2026.05 | 2.74 | 33.8 | 50 | 45 | |
| AC sweepModel=GPT-OSS-120B, Target accuracy (alpha)=75%2026.05 | 1.93 | 33.8 | 50 | 45 | |
| AC sweepModel=GPT-OSS-120B, Target accuracy (alpha)=90%2026.05 | 1.54 | 33.8 | 50 | 107 | |
| P(True)Model=GPT-OSS-120B, Target accuracy (alpha)=75%2026.05 | 1.23 | 33.8 | 50 | 45 | |
| P(True)Model=GPT-OSS-120B, Target accuracy (alpha)=90%2026.05 | 1.2 | 33.8 | 50 | 107 | |
| Self-certaintyModel=Nemotron3-30B, Target accuracy (alpha)=75%2026.05 | 1.04 | 29.5 | 49.3 | 486 | |
| Self-certaintyModel=Nemotron3-30B, Target accuracy (alpha)=90%2026.05 | 1.01 | 29.5 | 49.3 | 1.3 | |
| ESC sweepModel=GPT-OSS-120B, Target accuracy (alpha)=99%2026.05 | 0.94 | 33.8 | 50 | 2.5 | |
| DeepConf tailModel=GPT-OSS-120B, Target accuracy (alpha)=75%2026.05 | 0.91 | 33.8 | 50 | 45 | |
| DeepConf tailModel=GPT-OSS-120B, Target accuracy (alpha)=90%2026.05 | 0.91 | 33.8 | 50 | 107 | |
| Self-certaintyModel=GPT-OSS-120B, Target accuracy (alpha)=75%2026.05 | 0.89 | 33.8 | 50 | 45 | |
| Self-certaintyModel=GPT-OSS-120B, Target accuracy (alpha)=90%2026.05 | 0.87 | 33.8 | 50 | 107 | |
| Self-certaintyModel=Nemotron3-30B, Target accuracy (alpha)=99%2026.05 | 0.85 | 29.5 | 49.3 | 6 | |
| PC-linearModel=GPT-OSS-120B, Target accuracy (alpha)=75%2026.05 | 0.7 | 33.8 | 50 | 45 | |
| PC-quadraticModel=GPT-OSS-120B, Target accuracy (alpha)=75%2026.05 | 0.65 | 33.8 | 50 | 45 | |
| PC-cubicModel=GPT-OSS-120B, Target accuracy (alpha)=75%2026.05 | 0.65 | 33.8 | 50 | 45 | |
| PC-linearModel=GPT-OSS-120B, Target accuracy (alpha)=90%2026.05 | 0.63 | 33.8 | 50 | 107 | |
| PC-quadraticModel=GPT-OSS-120B, Target accuracy (alpha)=90%2026.05 | 0.52 | 33.8 | 50 | 107 | |
| PC-cubicModel=GPT-OSS-120B, Target accuracy (alpha)=90%2026.05 | 0.52 | 33.8 | 50 | 107 | |
| AC sweepModel=GPT-OSS-120B, Target accuracy (alpha)=99%2026.05 | 0.29 | 33.8 | 50 | 2.5 | |
| P(True)Model=GPT-OSS-120B, Target accuracy (alpha)=99%2026.05 | 0.23 | 33.8 | 50 | 2.5 | |
| PC-cubicModel=Nemotron3-30B, Target accuracy (alpha)=99%2026.05 | 0.22 | 29.5 | 49.3 | 6 | |
| PC-linearModel=Ministral3-14B, Target accuracy (alpha)=99%2026.05 | 0.1 | 9.1 | 14.2 | 924 | |
| PC-quadraticModel=GPT-OSS-120B, Target accuracy (alpha)=99%2026.05 | 0.06 | 33.8 | 50 | 2.5 | |
| PC-cubicModel=GPT-OSS-120B, Target accuracy (alpha)=99%2026.05 | 0.05 | 33.8 | 50 | 2.5 |