Human Preference Prediction on DocPairBench benchmark leaderboard · SOTA2 Research

Human Preference Prediction on DocPairBench

89.3Gov. Preference ScoreDocReward-7B

Evaluation Results

Method	Links
DocReward-7B2025.10		89.3	92.3	86.8	83.8	67	87	84.4	80.9	76.4	76.7	82.3
DocReward-3B2025.10		87.2	76.9	80.2	75	69	85	87.2	79.1	74.6	77.3	80.6
GPT-52025.10		70.2	69.2	69.2	68.8	57	69	76.2	63.6	70	69.8	68.9
GPT-52025.10		69.7	80.8	70.3	71.3	49	72	75.2	60	64.6	66.6	67.7
Claude Sonnet 42025.10		62.9	69.2	75.8	61.3	46	60	75.2	56.4	65.5	62.9	63
GPT-4o2025.10		61.9	57.7	75.8	66.3	53	62	71.6	61.8	66.4	61.4	63.3
GPT-4o2025.10		60.1	46.2	58.2	56.3	34	54	58.7	50.9	58.2	53.9	54.9
Claude Sonnet 42025.10		59.8	50	56	60	48	63	48.6	50	50.9	49.7	54.2
Qwen2.5-VL-7B2025.10		57.7	61.5	53.9	53.8	54	59	56	52.7	49.1	58.1	56.1
Qwen2.5-VL-3B2025.10		55.1	61.5	48.4	50	50	47	56.9	47.3	55.5	60.5	54.4
Qwen2.5-VL-7B2025.10		49.4	34.6	46.2	41.3	35	52	49.5	31.8	48.2	48.5	46
Qwen2.5-VL-3B2025.10		38.4	23.1	36.3	30	25	27	32.1	39.1	38.2	30.2	33.5