SOTA Math Reasoning on GSM8K (Accuracy and Token Delta) and PapersWithCode

96.7Accuracy

Qwen3-Next-80B

Updated 1mo ago

Evaluation Results

Method
Qwen3-Next-80B 2025.10	96.7	-
Qwen3-32B + THINKBRAKE 2025.10	96.5	-9.1
Qwen3-Next-80B + THINKBRAKE 2025.10	96.5	-24.3
Qwen3-14B 2025.10	96	-
Qwen3-32B 2025.10	96	-
Qwen3-14B + NoWait 2025.10	95.8	-29.4
Qwen3-32B + NoWait 2025.10	95.3	-23.6
Qwen3-4B-Thinking + Dynasor-CoT 2025.10	95.2	-34
Qwen3-4B-Thinking 2025.10	95.1	-
Qwen3-4B-Thinking + DEER 2025.10	95.1	-34.1
Qwen3-14B + THINKBRAKE-p 2025.10	95.1	-31.6
Qwen3-14B + THINKBRAKE 2025.10	95	-15.4
Qwen3-4B-Thinking + THINKBRAKE 2025.10	94.8	-18.9
Qwen3-32B + THINKBRAKE-p 2025.10	94.8	-23
Qwen3-4B + NoWait 2025.10	94.5	-40.4
Qwen3-4B + THINKBRAKE 2025.10	94.5	-30.1
Qwen3-4B 2025.10	94.4	-
Qwen3-4B-Thinking + NoWait 2025.10	94.2	-25.1
Qwen3-4B-Thinking + THINKBRAKE-p 2025.10	93.9	-37.3
Qwen3-14B + ThinkLess 2025.10	93.4	-100
Qwen3-32B + ThinkLess 2025.10	93.3	-100
Qwen3-4B + THINKBRAKE-p 2025.10	92.8	-40.2
DeepSeek-R1-7B 2025.10	92.7	-
Qwen3-4B-Thinking + ThinkLess 2025.10	92.6	-100
Phi-4-Reasoning + ThinkLess 2025.10	92.6	-100
DeepSeek-R1-7B + THINKBRAKE 2025.10	92.4	-25
Phi-4-Reasoning + NoWait 2025.10	92.1	-2.4
Phi-4-Reasoning 2025.10	91.7	-
Phi-4-Reasoning + THINKBRAKE 2025.10	91.7	-25.6
Phi-4-Reasoning + THINKBRAKE-p 2025.10	91.3	-42.5
Qwen3-4B + ThinkLess 2025.10	91	-100
DeepSeek-R1-7B + NoWait 2025.10	90.6	-30.8
GWO/n=6+CoT 2026.04	90.6	-
GWO+CoT 2026.04	89.7	-
GWO 2026.04	89.1	-
DeepSeek-R1-7B + DEER 2025.10	89	-60.5
OMI2 (Full) 2026.04	88.5	-
DeepSeek-R1-7B + THINKBRAKE-p 2025.10	87.8	-54.9
WI 2026.04	87	-
WI (Full) 2026.04	87	-
OC-SFT-1 2026.04	86.7	-
CODEI/O 2026.04	86.4	-
DeepSeek-R1-7B + ThinkLess 2025.10	85.9	-100
PyEdu 2026.04	85.8	-
CODEI/O++ 2026.04	85.7	-
OMI2 2026.04	84.1	-
CoT-SC/n=5 2026.04	80.1	-
CoT 2026.04	77.3	-
DeepSeek-R1-7B + Dynasor-CoT 2025.10	61.3	-87.8