Mathematical Reasoning on OlympiadBench (Accuracy, Thinking Token Usage (%))
71.5AccuracyAT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ATModel Scale=8B, Thinking Strategy=AT2026.06 | 71.5 | — | |
| DARTModel Scale=8B, Thinking Strategy=DART2026.06 | 69.8 | 45 | |
| DARTModel Scale=14B, Thinking Strategy=DART2026.06 | 62 | 15 | |
| DARTModel Scale=32B, Thinking Strategy=DART2026.06 | 58.5 | 16 | |
| ATModel Scale=32B, Thinking Strategy=AT2026.06 | 54 | — | |
| ATModel Scale=14B, Thinking Strategy=AT2026.06 | 53 | — | |
| NTModel Scale=14B, Thinking Strategy=NT2026.06 | 51.5 | — | |
| NTModel Scale=32B, Thinking Strategy=NT2026.06 | 50.5 | — | |
| NTModel Scale=8B, Thinking Strategy=NT2026.06 | 49.8 | — |