Visual Question Answering on GAR-Bench-VQA benchmark leaderboard · SOTA2 Research

Visual Question Answering on GAR-Bench-VQA

64.2Overall VQA ScoreGemini-2.5-Pro

Evaluation Results

Method	Links
Gemini-2.5-Pro2025.10		64.2	—	62.3	68.8	58.6	66.7	—	64.1	64.9	70.3
o32025.10		61.3	—	58	70.3	55.2	63.9	—	54.7	49.2	71.3
GAR-8B2025.10		59.9	—	59.4	54.7	75.9	52.8	—	48.4	60.7	68.3
GPT-4o2025.10		53.5	—	34.8	65.3	48.3	52.8	—	57.8	60.2	61.4
Qwen2.5-VL-72B2025.10		52.8	—	46.4	50	65.5	33.3	—	68.8	44.3	57.4
Qwen2.5-VL-32B2025.10		50.9	—	46.4	53.1	41.4	30.6	—	71.9	36.1	58.4
GAR-1B2025.10		50.6	—	55.1	46.9	69	47.2	—	21.9	62.3	56.4
InternVL3-78B2025.10		50.5	—	44.9	54.7	58.6	61.1	—	53.1	47.5	45.5
InternVL3-38B2025.10		46.5	—	39.1	40.6	51.7	55.6	—	60.9	36.1	47.5
Qwen2.5-VL-7B2025.10		41.7	—	39.1	40.6	44.8	27.8	—	59.4	36.1	40.6
InternVL3-8B2025.10		38.9	—	36.2	37.5	58.6	41.7	—	51.6	27.9	33.6
DAM-3B2025.10		38.2	—	55.1	39.1	41.4	36.1	—	31.3	36.1	31.7
VP-SPHINX-13B2025.10		37.5	—	33.3	25	44.8	38.9	—	60.9	34.3	32.7
InternVL3-2B2025.10		35.1	—	30.4	21.9	48.3	38.9	—	48.4	26.2	38.6
Qwen2.5-VL-3B2025.10		34.4	—	29	25	34.5	30.6	—	43.8	26.2	44.6
Sa2VA-8B2025.10		34.3	—	39.1	45.3	29.6	30.6	—	54.7	21.3	21.8
PAM-3B2025.10		2.4	—	2.9	3.1	6.9	5.6	—	1.6	1.6	0