Mathematical and Scientific Reasoning on Average (GSM8K, MATH-500, AIME 2024, OlympiadBench, GPQA Diamond)
78.4AccuracyASAG
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ASAGBackbone=Qwen3-14B2026.06 | 78.4 | 68.4 | |
| DEERBackbone=Qwen3-14B2026.06 | 77.2 | 68.2 | |
| DynasorBackbone=Qwen3-14B2026.06 | 76.4 | 82.3 | |
| VanillaBackbone=Qwen3-14B2026.06 | 75.8 | 100 | |
| ASAGBackbone=Qwen3-8B2026.06 | 75.1 | 64.2 | |
| TALEBackbone=Qwen3-14B2026.06 | 75.1 | 75.3 | |
| ASAGBackbone=Qwen3-4B2026.06 | 73.9 | 63.5 | |
| DEERBackbone=Qwen3-8B2026.06 | 73.4 | 63.5 | |
| DynasorBackbone=Qwen3-8B2026.06 | 72.8 | 69.7 | |
| VanillaBackbone=Qwen3-8B2026.06 | 72.5 | 100 | |
| DynasorBackbone=Qwen3-4B2026.06 | 71.6 | 64.4 | |
| DEERBackbone=Qwen3-4B2026.06 | 71.3 | 64.4 | |
| VanillaBackbone=Qwen3-4B2026.06 | 71 | 100 | |
| TALEBackbone=Qwen3-8B2026.06 | 70.5 | 68.9 | |
| TALEBackbone=Qwen3-4B2026.06 | 67.1 | 58.7 | |
| NoThinkingBackbone=Qwen3-14B2026.06 | 62.5 | 43.7 | |
| ASAGBackbone=DeepSeek-R1-Distill-Qwen-7B2026.06 | 62.3 | 64.7 | |
| DEERBackbone=DeepSeek-R1-Distill-Qwen-7B2026.06 | 60.5 | 63.1 | |
| NoThinkingBackbone=Qwen3-8B2026.06 | 59.8 | 40.3 | |
| DynasorBackbone=DeepSeek-R1-Distill-Qwen-7B2026.06 | 59.7 | 78 | |
| VanillaBackbone=DeepSeek-R1-Distill-Qwen-7B2026.06 | 58.2 | 100 | |
| TALEBackbone=DeepSeek-R1-Distill-Qwen-7B2026.06 | 57.1 | 67.5 | |
| NoThinkingBackbone=Qwen3-4B2026.06 | 54.8 | 34.3 | |
| NoThinkingBackbone=DeepSeek-R1-Distill-Qwen-7B2026.06 | 49.4 | 31.8 |