Commonsense Reasoning on PIQA, ARC-e, ARC-c, HellaSwag, and WinoGrande benchmark leaderboard · SOTA2 Research

Commonsense Reasoning on PIQA, ARC-e, ARC-c, HellaSwag, and WinoGrande

74.1PIQA AccuracyBitNetV2-7B

Evaluation Results

Method	Links
BitNetV2-7B2026.06		74.1	58.54	32.94	61.08	61.48	57.63
FlashMorph2026.06		73.3	74.5	44.7	61.2	62.3	63.2
FlashMorph2026.06		73.3	74.4	44.1	61.5	63.9	63.4
FlashMorph2026.06		73.2	73.2	42.3	60.6	61	62.1
PostNAS*2026.06		73.2	73.4	43.1	61.5	60.1	62.3
HALO2026.06		73.2	75.8	45.7	61.1	63.1	63.8
PostNAS*2026.06		73.1	75.1	45.1	61.7	62.3	63.4
PostNAS*2026.06		72.9	74.7	43	60.8	60.9	62.5
KL-LS2026.06		72.9	74.3	47.6	61.6	62.3	63.7
Uniform2026.06		72.7	74.4	45.5	60.8	63	63.3
HALO2026.06		72.6	73.5	42.8	60.6	61.5	62.2
HALO2026.06		72.6	74.4	44.5	60.9	62.8	63
Qwen32026.06		72.5	72.4	43.5	60.4	61	62
KL-LS2026.06		72.4	74.6	46.3	61	59.8	62.8
Uniform2026.06		72.3	74.7	45.1	60.5	62.2	62.9
KL-LS2026.06		72.3	74.2	42.8	60.4	59.1	61.8
Uniform2026.06		72.3	74.3	44.5	61.5	62.4	63
Qwen3-8B + CAT-Q2026.06		72.09	68.94	42.49	58.64	62.67	60.96
BitNetV2-3B2026.06		71.33	55.56	30.55	57.19	58.72	54.67
BitNetV2-1.3B2026.06		69.42	49.96	27.9	48.37	57.22	50.57
Qwen3-4B + CAT-Q2026.06		68.65	62.64	35.69	51.48	63.51	56.39
Qwen3-1.7B + CAT-Q2026.06		68.16	54.94	28.67	47.22	54.62	50.72
KL-LS2026.06		67.9	61.1	35.3	47.4	56.7	53.7
FlashMorph2026.06		67.8	63.1	33.5	47.5	56.4	53.7
HALO2026.06		67.7	60.1	34.1	47.5	55.3	53
Qwen32026.06		67.6	60.6	34.1	47.3	55.7	53.1
KL-LS2026.06		67.5	62.1	33	46.2	54.9	52.7
FlashMorph2026.06		67.5	63.9	32.9	46.9	54.1	53.1
HALO2026.06		67.4	63.5	32.4	46.4	56.1	53.2
HALO2026.06		67.4	62.9	32.6	46.9	55.1	53
FlashMorph2026.06		67.3	62.8	32	46.4	55.2	52.7
Uniform2026.06		67.3	63.1	33.1	46.8	55.3	53.1
KL-LS2026.06		67.2	61.8	33.8	46.9	54.9	52.9
Uniform2026.06		67.1	59.6	32.3	47.6	55.3	52.4
Uniform2026.06		66.9	62.3	32.9	46.1	55.6	52.8