Multimodal Reasoning Evaluation on CRYSTAL benchmark leaderboard · SOTA2 Research

Multimodal Reasoning Evaluation on CRYSTAL

57.99AccuracyGPT-5

Evaluation Results

Method	Links
GPT-52026.03		57.99	61.2	92.5	47.9	5.29	63.6	53.9
Qwen3-VL-8B2026.03		57.66	65.9	82.7	59	7.37	62.4	57.2
GPT-5-mini2026.03		55.59	77.3	97.8	66.9	7.57	56	67
Gemini 2.5 Flash2026.03		53.95	67.3	70.1	76.5	17.1	58.4	57.9
InternVL3.5-8B2026.03		51.98	53	88.2	41.6	4.96	69.2	46.9
InternVL3.5-38B2026.03		51.21	61.2	89.2	49.8	5.76	64.3	53.8
Qwen3-VL-32B2026.03		49.22	71.8	81.9	70.4	10.56	58.1	61.7
Qwen2.5-VL-32B2026.03		47.63	65.3	94.3	52.4	5.86	61.9	57.2
GPT-5.2 Instant2026.03		47.35	56.4	97.4	41.6	4.64	64.8	50.1
Qwen2.5-VL-3B2026.03		39.85	48	89.8	34.7	3.73	72.3	43.4
InternVL3.5-4B2026.03		37.61	43.2	89.5	32.5	3.75	77.5	38.7
Qwen3-VL-2B2026.03		34.15	59.5	72.6	53.5	5.94	66.9	51.5
Gemma3-12B2026.03		33.83	60.5	83.8	49.9	5.51	67.3	53.4
InternVL3.5-2B2026.03		33.02	46.9	72.5	37.1	3.92	73.1	41.5
Qwen2.5-VL-7B2026.03		30.43	47.5	76.5	36.5	4.07	71.7	42.2
InternVL3.5-1B2026.03		30.13	33	61.6	24.3	2.51	80.7	29.7
Gemma3-4B2026.03		28.65	61.8	87.8	50.6	5.72	66.8	54.7
MiniCPMv2.6-8B2026.03		25.54	21.5	70.9	13.4	1.31	85.4	18.6
Llama 3.2-11B2026.03		24.83	47.1	71.3	37.9	4.19	72.6	41.5
LLaVA-v1.6-7B2026.03		24.66	51.2	96.1	37	3.94	67.5	45.9