Mathematical Reasoning on GSM8K (Acc, Tok, CR, Cost)
96.7AccuracyVanilla
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| VanillaModel=Qwen3-14B2026.04 | 96.7 | 1,668 | 100 | 1,668 | |
| EATModel=Qwen3-14B2026.04 | 96.7 | 1,591 | 95.4 | 1,639 | |
| RCPDModel=Qwen3-14B2026.04 | 96.7 | 1,394 | 83.6 | 1,476 | |
| DEERModel=Qwen3-14B2026.04 | 96.4 | 763 | 45.7 | 778 | |
| CoDE-StopModel=Qwen3-14B2026.04 | 95.2 | 695 | 41.7 | 709 | |
| VanillaModel=Qwen3-4B2026.04 | 94.8 | 2,306 | 100 | 2,306 | |
| DEERModel=Qwen3-4B2026.04 | 94.8 | 1,374 | 59.6 | 1,400 | |
| RCPDModel=Qwen3-4B2026.04 | 94.7 | 2,140 | 92.8 | 2,264 | |
| EATModel=Qwen3-4B2026.04 | 94.6 | 2,092 | 90.7 | 2,161 | |
| CoDE-StopModel=Qwen3-4B2026.04 | 94.6 | 1,208 | 52.4 | 1,233 | |
| VanillaModel=Nemotron-8B2026.04 | 91.7 | 1,211 | 100 | 1,211 | |
| RCPDModel=Nemotron-8B2026.04 | 91.7 | 1,182 | 97.6 | 1,253 | |
| Think or NotModel=Nemotron-8B2026.04 | 91.6 | 1,162 | 96 | 1,724 | |
| EATModel=Nemotron-8B2026.04 | 91.6 | 1,194 | 98.6 | 1,229 | |
| DEERModel=Nemotron-8B2026.04 | 91.1 | 984 | 81.3 | 1,007 | |
| RCPDModel=DeepSeek-R12026.04 | 90.8 | 1,840 | 100 | 1,934 | |
| CoDE-StopModel=Nemotron-8B2026.04 | 90.8 | 860 | 71 | 881 | |
| VanillaModel=DeepSeek-R12026.04 | 90.7 | 1,840 | 100 | 1,840 | |
| EATModel=DeepSeek-R12026.04 | 90.7 | 1,789 | 97.2 | 1,870 | |
| Think or NotModel=DeepSeek-R12026.04 | 90.6 | 1,808 | 98.3 | 3,319 | |
| Answer ConvergenceModel=Qwen3-14B2026.04 | 90.1 | 385 | 23.1 | 811 | |
| DEERModel=DeepSeek-R12026.04 | 89.5 | 1,254 | 68.2 | 1,322 | |
| CoDE-StopModel=DeepSeek-R12026.04 | 88.6 | 1,056 | 57.4 | 1,110 | |
| Think or NotModel=Qwen3-4B2026.04 | 87.4 | 1,618 | 70.2 | 2,462 | |
| Answer ConvergenceModel=Nemotron-8B2026.04 | 82.6 | 421 | 34.8 | 666 | |
| Answer ConvergenceModel=DeepSeek-R12026.04 | 80.2 | 518 | 28.2 | 827 | |
| Answer ConvergenceModel=Qwen3-4B2026.04 | 75.1 | 401 | 17.4 | 825 |