Mathematical Reasoning on GSM8K (test) (Accuracy, Average, Improvement (%)) benchmark leaderboard · SOTA2 Research

Mathematical Reasoning on GSM8K (test) (Accuracy, Average, Improvement (%))

0.8385AccuracyT3D

Evaluation Results

Method	Links
T3D2026.02		0.8385	0.5889	12.43
T3D2026.02		0.8196	0.5889	12.43
Original Model2026.02		0.7877	0.5238	—
SFT2026.02		0.7741	0.4676	-10.73
dParallel2026.02		0.7657	0.4583	-12.51
dParallel2026.02		0.7597	0.4583	-12.51
ReDi2026.02		0.7362	0.368	-29.74
Naive TD2026.02		0.7278	0.4082	-22.07
Original Model2026.02		0.7233	0.5238	—
ReDi2026.02		0.7187	0.368	-29.74
T3D2026.02		0.7096	0.4306	7.59
T3D2026.02		0.699	0.3966	85.02
T3D2026.02		0.6884	0.4306	7.59
dParallel2026.02		0.6823	0.4014	0.29
Naive TD2026.02		0.6801	0.4082	-22.07
dParallel2026.02		0.677	0.4014	0.29
Naive TD2026.02		0.6619	0.3605	-9.94
ReDi2026.02		0.6399	0.3238	-19.11
Naive TD2026.02		0.6399	0.3605	-9.94
T3D2026.02		0.6399	0.3966	85.02
Original Model2026.02		0.63	0.4003	—
SFT2026.02		0.6255	0.4174	4.28
ReDi2026.02		0.6217	0.3238	-19.11
SFT2026.02		0.6179	0.4174	4.28
Naive TD2026.02		0.5792	0.2892	34.95
Original Model2026.02		0.5588	0.4003	—
SFT2026.02		0.5542	0.285	32.98
SFT2026.02		0.546	0.4676	-10.73
dParallel2026.02		0.5383	0.298	39.05
ReDi2026.02		0.533	0.2157	0.65
Naive TD2026.02		0.508	0.2892	34.95
SFT2026.02		0.4814	0.285	32.98
ReDi2026.02		0.4784	0.2157	0.65
Naive TD2026.02		0.4602	0.2207	122.78
dParallel2026.02		0.4594	0.298	39.05
dParallel2026.02		0.4526	0.226	128.09
dParallel2026.02		0.4291	0.226	128.09
T3D2026.02		0.4291	0.2227	124.79
Original Model2026.02		0.4109	0.2143	—
Original Model2026.02		0.4102	0.2143	—
Naive TD2026.02		0.3973	0.2207	122.78
SFT2026.02		0.3965	0.1784	80.05
T3D2026.02		0.3738	0.2227	124.79
SFT2026.02		0.3662	0.1784	80.05
ReDi2026.02		0.3245	0.1347	35.95
ReDi2026.02		0.2972	0.1347	35.95
Original Model2026.02		0.1334	0.0991	—
Original Model2026.02		0.1274	0.0991	—