Mathematical Reasoning on MATH 500 (Accuracy, Token count)
95.4AccuracyVanilla
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VanillaBackbone=Qwen3-14B2026.04 | 95.4 | 4,503 | |
| DTSRBackbone=Qwen3-14B2026.04 | 95 | 2,247 | |
| DTSRBackbone=Qwen3-32B2026.04 | 95 | 2,660 | |
| VanillaBackbone=Qwen3-32B2026.04 | 94.8 | 4,489 | |
| VanillaBackbone=Qwen3-8B2026.04 | 94.4 | 5,067 | |
| DEERBackbone=Qwen3-14B2026.04 | 94.4 | 2,601 | |
| DTSRBackbone=Qwen3-8B2026.04 | 94 | 2,359 | |
| DEERBackbone=Qwen3-32B2026.04 | 94 | 2,881 | |
| DEERBackbone=Qwen3-8B2026.04 | 93.4 | 2,549 | |
| NoWAITBackbone=Qwen3-32B2026.04 | 93.4 | 3,285 | |
| NoWAITBackbone=Qwen3-8B2026.04 | 92.8 | 3,485 | |
| NoWAITBackbone=Qwen3-14B2026.04 | 92.8 | 3,307 | |
| NoThinkingBackbone=Qwen3-14B2026.04 | 88.2 | 852 | |
| NoThinkingBackbone=Qwen3-32B2026.04 | 86.4 | 958 | |
| NoThinkingBackbone=Qwen3-8B2026.04 | 86 | 1,067 |