Mathematical Reasoning on MATH 500 (Acc, Tok, CR)
95.2AccuracyVanilla
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| VanillaBackbone=Qwen3-8B2026.04 | 95.2 | 5,138 | 100 | |
| TRACEBackbone=Qwen3-8B2026.04 | 94.1 | 3,231 | 62.9 | |
| DynasorBackbone=Qwen3-8B2026.04 | 93.7 | 3,873 | 75.4 | |
| VanillaBackbone=Qwen3-4B2026.04 | 93.6 | 5,229 | 100 | |
| TRACEBackbone=Qwen3-4B2026.04 | 93.2 | 3,788 | 72.4 | |
| TALEBackbone=Qwen3-8B2026.04 | 93.1 | 3,544 | 69 | |
| DEERBackbone=Qwen3-8B2026.04 | 92.9 | 3,468 | 67.5 | |
| DEERBackbone=Qwen3-4B2026.04 | 92.8 | 3,944 | 75.4 | |
| VanillaBackbone=Gemini2.5-Flash2026.04 | 92.8 | 1,449 | 100 | |
| TRACEBackbone=Gemini2.5-Flash2026.04 | 92.2 | 1,166 | 80.5 | |
| DynasorBackbone=Gemini2.5-Flash2026.04 | 92 | 1,311 | 90.5 | |
| TALEBackbone=Gemini2.5-Flash2026.04 | 91.9 | 1,411 | 97.4 | |
| DynasorBackbone=Qwen3-4B2026.04 | 91.4 | 4,339 | 83 | |
| DEERBackbone=Gemini2.5-Flash2026.04 | 91.2 | 1,224 | 84.5 | |
| NoThinkBackbone=Gemini2.5-Flash2026.04 | 90.6 | 1,400 | 96.6 | |
| TALEBackbone=Qwen3-4B2026.04 | 89.7 | 2,938 | 56.2 | |
| VanillaBackbone=R1-Distilled-Llama-8B2026.04 | 87.2 | 2,934 | 100 | |
| DynasorBackbone=R1-Distilled-Llama-8B2026.04 | 86.4 | 2,693 | 91.8 | |
| DEERBackbone=R1-Distilled-Llama-8B2026.04 | 86.4 | 2,718 | 92.6 | |
| TRACEBackbone=R1-Distilled-Llama-8B2026.04 | 85.4 | 2,436 | 83 | |
| TALEBackbone=R1-Distilled-Llama-8B2026.04 | 85.3 | 2,455 | 83.7 | |
| NoThinkBackbone=R1-Distilled-Llama-8B2026.04 | 85.2 | 3,024 | 103 | |
| NoThinkBackbone=Qwen3-8B2026.04 | 84.2 | 923 | 18 | |
| NoThinkBackbone=Qwen3-4B2026.04 | 80.4 | 908 | 17.4 |