Ride-hailing adjudication on Passenger-Cancel benchmark leaderboard · SOTA2 Research

Ride-hailing adjudication on Passenger-Cancel

95.7Normal PrecisionQwen3-VL-8B-Instruct

Evaluation Results

Method	Links
Qwen3-VL-8B-Instruct2026.03		95.7	13.8	91.5	16.3	75	62.09
Qwen3-VL-8B-Instruct2026.03		92.9	8	92.3	9.1	73.6	60.61
RideJudge-8B2026.03		89.76	80.92	86.48	85.71	87.4	88.41
DeepSeek-V3.12026.03		88.57	66.77	89.12	65.4	85.3	75.25
DeepSeek-V3.12026.03		86.75	66.46	88.38	66.54	85.3	75.36
Qwen3-VL-32B-Instruct2026.03		84.6	37.2	81.5	38.4	78.9	65.55
RideJudge-4B2026.03		84.5	85.54	87.69	86.69	86.63	87.25
Qwen3-VL-32B-Instruct2026.03		83.6	50.2	81.3	52.9	68.5	59.14
Qwen3-VL-8B-Thinking2026.03		66.7	1.2	33.3	0.4	8.7	7.37
QwQ-32B2026.03		38.03	91.38	53.71	54.37	18.51	35.18
MiniCPM-V2026.03		36.84	4.31	66.67	1.52	18.66	19.68
MiniCPM-V2026.03		35.85	5.85	45.45	1.9	8.71	12.96
QwQ-32B2026.03		32.47	46.46	53.91	23.57	19.44	35.82
Qwen3-32B-Instruct2026.03		30.73	41.23	53.98	23.19	17.26	26.77
Qwen3-32B-Instruct2026.03		30.33	39.38	53.66	16.73	16.72	26.21
Qwen3-VL-32B-Thinking2026.03		27.59	51.69	0	0	39.74	45.94
Qwen3-VL-32B-Thinking2026.03		27.52	27.69	0	0	54.98	56.86
Qwen3-VL-8B-Thinking2026.03		0	0	0	0	11.6	8.82