Reasoning and Code Generation on Reasoning Evaluation Suite (GSM8K, MATH500, AIME24, HumanEval) (test) benchmark leaderboard · SOTA2 Research

Reasoning and Code Generation on Reasoning Evaluation Suite (GSM8K, MATH500, AIME24, HumanEval) (test)

94.7GSM8K AccuracySafeChain

Evaluation Results

Method	Links
SafeChain2026.04		94.7	84.6	26.7	66.5	68.1
No train2026.04		93.7	80.4	20	69.5	65.9
STAR-12026.04		92.9	88.6	56.7	79.9	79.5
S1-ALT2026.04		92.7	80.6	20	65.8	64.8
SafeChain2026.04		92.1	83.8	50	81.7	76.9
R1-ALT2026.04		91.4	84.2	53.3	82.9	78
No train2026.04		91.2	85.6	46.7	80.5	76
STAR-12026.04		90.9	84.8	40	83.5	74.8
SafeChain2026.04		90.8	83.8	60	82.3	79.2
DirectRefusal2026.04		90.4	80.8	46.7	84.8	75.7
No train2026.04		89.9	84	40	83.5	74.4
DirectRefusal2026.04		89.8	80	36.7	81.1	71.9
SafeChain2026.04		89.2	83	40	82.3	73.6
SafeChain2026.04		89.1	83	36.7	81.7	72.6
R1-ALT2026.04		88.6	84.8	40	84.8	74.6
R1-ALT2026.04		86.6	84.6	36.7	70.1	69.5
SafeChain2026.04		86	80.6	16.7	64.6	62
No train2026.04		85.1	84.6	43.3	77.4	72.6
SafeChain2026.04		85.1	84.4	30	68.9	67.1
STAR-12026.04		85.1	85.6	36.7	77.4	71.2
DirectRefusal2026.04		80	75.4	36.7	31.1	55.8
S1-ALT2026.04		79.8	51	0	38.4	42.3
No train2026.04		78.6	53.2	0	29.9	40.4
SafeChain2026.04		77.7	49.4	0	39.6	41.7
SafeChain2026.04		72	71.6	16.7	66.5	56.7
SafeChain2026.04		70.7	76.6	30	67.1	61.1
No train2026.04		70.2	72.4	23.3	66.5	58.1
STAR-12026.04		69.6	69.8	16.7	67.7	56
R1-ALT2026.04		69	74.4	26.7	68.9	59.8
DirectRefusal2026.04		66.9	65	20	64.6	54.1
SafeChain2026.04		51.4	45.2	0	43.9	35.1
No train2026.04		50.3	44.6	6.7	42.7	36.1
SafeChain2026.04		49.7	46	0	45.7	35.4
R1-ALT2026.04		49.4	43.6	13.3	39	36.3
DirectRefusal2026.04		49.3	45.2	3.3	43.9	35.4
STAR-12026.04		45	51.2	10	53.7	40