Math Reasoning on GSM8K (Accuracy and Token Delta)
96.7AccuracyQwen3-Next-80B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen3-Next-80BBackbone=Qwen3-Next-80B, Decoding Strategy=Base2025.10 | 96.7 | — | |
| Qwen3-32B + THINKBRAKEBackbone=Qwen3-32B, Decoding Strategy=THINKBRAKE2025.10 | 96.5 | -9.1 | |
| Qwen3-Next-80B + THINKBRAKEBackbone=Qwen3-Next-80B, Decoding Strategy=THINKBRAKE2025.10 | 96.5 | -24.3 | |
| Qwen3-14BBackbone=Qwen3-14B, Decoding Strategy=Base2025.10 | 96 | — | |
| Qwen3-32BBackbone=Qwen3-32B, Decoding Strategy=Base2025.10 | 96 | — | |
| Qwen3-14B + NoWaitBackbone=Qwen3-14B, Decoding Strategy=NoWait2025.10 | 95.8 | -29.4 | |
| Qwen3-32B + NoWaitBackbone=Qwen3-32B, Decoding Strategy=NoWait2025.10 | 95.3 | -23.6 | |
| Qwen3-4B-Thinking + Dynasor-CoTBackbone=Qwen3-4B-Thinking, Decoding Strategy=Dynasor-CoT2025.10 | 95.2 | -34 | |
| Qwen3-4B-ThinkingBackbone=Qwen3-4B-Thinking, Decoding Strategy=Base2025.10 | 95.1 | — | |
| Qwen3-4B-Thinking + DEERBackbone=Qwen3-4B-Thinking, Decoding Strategy=DEER2025.10 | 95.1 | -34.1 | |
| Qwen3-14B + THINKBRAKE-pBackbone=Qwen3-14B, Decoding Strategy=THINKBRAKE-p2025.10 | 95.1 | -31.6 | |
| Qwen3-14B + THINKBRAKEBackbone=Qwen3-14B, Decoding Strategy=THINKBRAKE2025.10 | 95 | -15.4 | |
| Qwen3-4B-Thinking + THINKBRAKEBackbone=Qwen3-4B-Thinking, Decoding Strategy=THINKBRAKE2025.10 | 94.8 | -18.9 | |
| Qwen3-32B + THINKBRAKE-pBackbone=Qwen3-32B, Decoding Strategy=THINKBRAKE-p2025.10 | 94.8 | -23 | |
| Qwen3-4B + NoWaitBackbone=Qwen3-4B, Decoding Strategy=NoWait2025.10 | 94.5 | -40.4 | |
| Qwen3-4B + THINKBRAKEBackbone=Qwen3-4B, Decoding Strategy=THINKBRAKE2025.10 | 94.5 | -30.1 | |
| Qwen3-4BBackbone=Qwen3-4B, Decoding Strategy=Base2025.10 | 94.4 | — | |
| Qwen3-4B-Thinking + NoWaitBackbone=Qwen3-4B-Thinking, Decoding Strategy=NoWait2025.10 | 94.2 | -25.1 | |
| Qwen3-4B-Thinking + THINKBRAKE-pBackbone=Qwen3-4B-Thinking, Decoding Strategy=THINKBRAKE-p2025.10 | 93.9 | -37.3 | |
| Qwen3-14B + ThinkLessBackbone=Qwen3-14B, Decoding Strategy=ThinkLess2025.10 | 93.4 | -100 | |
| Qwen3-32B + ThinkLessBackbone=Qwen3-32B, Decoding Strategy=ThinkLess2025.10 | 93.3 | -100 | |
| Qwen3-4B + THINKBRAKE-pBackbone=Qwen3-4B, Decoding Strategy=THINKBRAKE-p2025.10 | 92.8 | -40.2 | |
| DeepSeek-R1-7BBackbone=DeepSeek-R1-7B, Decoding Strategy=Base2025.10 | 92.7 | — | |
| Qwen3-4B-Thinking + ThinkLessBackbone=Qwen3-4B-Thinking, Decoding Strategy=ThinkLess2025.10 | 92.6 | -100 | |
| Phi-4-Reasoning + ThinkLessBackbone=Phi-4-Reasoning, Decoding Strategy=ThinkLess2025.10 | 92.6 | -100 | |
| DeepSeek-R1-7B + THINKBRAKEBackbone=DeepSeek-R1-7B, Decoding Strategy=THINKBRAKE2025.10 | 92.4 | -25 | |
| Phi-4-Reasoning + NoWaitBackbone=Phi-4-Reasoning, Decoding Strategy=NoWait2025.10 | 92.1 | -2.4 | |
| Phi-4-ReasoningBackbone=Phi-4-Reasoning, Decoding Strategy=Base2025.10 | 91.7 | — | |
| Phi-4-Reasoning + THINKBRAKEBackbone=Phi-4-Reasoning, Decoding Strategy=THINKBRAKE2025.10 | 91.7 | -25.6 | |
| Phi-4-Reasoning + THINKBRAKE-pBackbone=Phi-4-Reasoning, Decoding Strategy=THINKBRAKE-p2025.10 | 91.3 | -42.5 | |
| Qwen3-4B + ThinkLessBackbone=Qwen3-4B, Decoding Strategy=ThinkLess2025.10 | 91 | -100 | |
| DeepSeek-R1-7B + NoWaitBackbone=DeepSeek-R1-7B, Decoding Strategy=NoWait2025.10 | 90.6 | -30.8 | |
| GWO/n=6+CoTBase=Qwen2.52026.04 | 90.6 | — | |
| GWO+CoTBase=Qwen2.52026.04 | 89.7 | — | |
| GWOBase=Qwen2.52026.04 | 89.1 | — | |
| DeepSeek-R1-7B + DEERBackbone=DeepSeek-R1-7B, Decoding Strategy=DEER2025.10 | 89 | -60.5 | |
| OMI2 (Full)Base=Qwen2.52026.04 | 88.5 | — | |
| DeepSeek-R1-7B + THINKBRAKE-pBackbone=DeepSeek-R1-7B, Decoding Strategy=THINKBRAKE-p2025.10 | 87.8 | -54.9 | |
| WIBase=Qwen2.52026.04 | 87 | — | |
| WI (Full)Base=Qwen2.52026.04 | 87 | — | |
| OC-SFT-1Base=Qwen2.52026.04 | 86.7 | — | |
| CODEI/OBase=Qwen2.52026.04 | 86.4 | — | |
| DeepSeek-R1-7B + ThinkLessBackbone=DeepSeek-R1-7B, Decoding Strategy=ThinkLess2025.10 | 85.9 | -100 | |
| PyEduBase=Qwen2.52026.04 | 85.8 | — | |
| CODEI/O++Base=Qwen2.52026.04 | 85.7 | — | |
| OMI2Base=Qwen2.52026.04 | 84.1 | — | |
| CoT-SC/n=5Base=Qwen2.52026.04 | 80.1 | — | |
| CoTBase=Qwen2.52026.04 | 77.3 | — | |
| DeepSeek-R1-7B + Dynasor-CoTBackbone=DeepSeek-R1-7B, Decoding Strategy=Dynasor-CoT2025.10 | 61.3 | -87.8 |