Mathematical Reasoning on OlympiadBench (Acc, Tok, CR)
81AccuracyVanilla
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| VanillaBackbone=Qwen3-8B2026.04 | 81 | 10,518 | 100 | |
| TRACEBackbone=Qwen3-8B2026.04 | 80 | 8,088 | 76.9 | |
| VanillaBackbone=Qwen3-4B2026.04 | 77.9 | 10,294 | 100 | |
| TRACEBackbone=Qwen3-4B2026.04 | 77.6 | 9,489 | 92.2 | |
| VanillaBackbone=Gemini2.5-Flash2026.04 | 77.6 | 6,684 | 100 | |
| DynasorBackbone=Qwen3-8B2026.04 | 77.3 | 9,148 | 87 | |
| TRACEBackbone=Gemini2.5-Flash2026.04 | 75.3 | 4,291 | 64.2 | |
| DEERBackbone=Qwen3-4B2026.04 | 74.7 | 9,696 | 94.2 | |
| DEERBackbone=Qwen3-8B2026.04 | 74.2 | 8,074 | 76.8 | |
| NoThinkBackbone=Gemini2.5-Flash2026.04 | 74.1 | 6,296 | 94.2 | |
| DynasorBackbone=Qwen3-4B2026.04 | 73.8 | 9,133 | 88.7 | |
| TALEBackbone=Qwen3-8B2026.04 | 73.1 | 8,417 | 80 | |
| DynasorBackbone=Gemini2.5-Flash2026.04 | 72.7 | 5,919 | 88.5 | |
| DEERBackbone=Gemini2.5-Flash2026.04 | 70.3 | 5,805 | 86.8 | |
| TALEBackbone=Qwen3-4B2026.04 | 69.5 | 8,521 | 82.8 | |
| TALEBackbone=Gemini2.5-Flash2026.04 | 68.4 | 4,950 | 74 | |
| VanillaBackbone=R1-Distilled-Llama-8B2026.04 | 58.2 | 6,286 | 100 | |
| TRACEBackbone=R1-Distilled-Llama-8B2026.04 | 57.5 | 4,784 | 76.1 | |
| NoThinkBackbone=Qwen3-8B2026.04 | 57.2 | 1,710 | 16.3 | |
| DynasorBackbone=R1-Distilled-Llama-8B2026.04 | 56.9 | 5,814 | 92.5 | |
| TALEBackbone=R1-Distilled-Llama-8B2026.04 | 56.2 | 5,713 | 90.9 | |
| NoThinkBackbone=R1-Distilled-Llama-8B2026.04 | 56 | 6,150 | 97.8 | |
| DEERBackbone=R1-Distilled-Llama-8B2026.04 | 55.4 | 5,443 | 86.6 | |
| NoThinkBackbone=Qwen3-4B2026.04 | 55 | 1,804 | 17.5 |