Faithful Perception on SPD-Faith Bench Multi-Difference Subset 1.0 (test) benchmark leaderboard · SOTA2 Research

Faithful Perception on SPD-Faith Bench Multi-Difference Subset 1.0 (test)

97.5Precision (Type: Color)InternVL2.5-38B

Evaluation Results

Method	Links
InternVL2.5-38B2026.02		97.5	51.4	67.3	93.3	31.2	46.8	52.8	69.2	59.9	81.2	50.6	58	74.8	45.6	56.7
DeepSeek-VL22026.02		96.8	32.9	49.1	94.6	36.4	52.6	62.1	79.2	69.6	84.5	49.5	57.1	29.2	18.7	22.8
InternVL2.5-8B2026.02		95.7	23.1	37.2	91.1	21.2	34.4	46.3	74.3	57	77.7	39.5	42.9	56.7	29.3	38.6
Qwen3-VL-235B-A22B2026.02		95.3	47.1	63.1	95.4	35.5	51.8	56.4	72.7	63.5	82.4	51.8	59.5	81.2	48.8	60.9
Gemini-2.5-Pro2026.02		93.5	59	72.4	94.3	38.2	54.3	89.3	59.7	71.6	92.4	52.3	66.1	88.2	48.7	62.8
Qwen2.5-VL-72B2026.02		93.3	47.5	63	97	32.9	49.2	56.2	59.2	57.7	82.2	46.5	56.6	75.1	44.1	55.5
Qwen2.5-VL-7B2026.02		93.3	33.2	48.9	95.2	34.6	50.7	46.2	68.9	55.3	78.2	45.6	51.6	50.1	31	38.3
Claude-4.5-Haiku2026.02		92.1	54.8	68.7	98.1	23.8	38.3	69.3	48.4	57	86.5	42.3	54.7	75.2	39.1	51.4
Qwen3-VL-32B2026.02		91.3	61.2	73.3	98.7	33.5	50	70.2	53	60.4	86.7	49.2	61.2	85.4	49.8	62.3
GLM-4.5V2026.02		90.4	66.5	76.7	95.4	50.4	66.1	91.8	50.8	65.4	92.5	55.9	69.4	88.9	57.2	69.6
GPT-4o2026.02		88.6	55.2	68	94.4	32.1	48	67.8	59.1	63.2	83.6	48.8	59.7	74.3	44.4	55.6
MiniCPM-V-2.62026.02		86.7	20.2	32.8	90.2	29.4	44.3	30.9	64.8	41.8	69.3	38.1	39.6	37.4	20.1	26.1