Multimodal NP-hard combinatorial optimization on MM-NP-Bench benchmark leaderboard · SOTA2 Research

Multimodal NP-hard combinatorial optimization on MM-NP-Bench

62.5Constraint Success RateDMPO-VL-7B

Evaluation Results

Method	Links
DMPO-VL-7B2026.05		62.5	31.5	25.5	22.1	98.5	73.1	72	67.4	51	21.6	61.9	43.1
Gemini-3.0-Flash2026.05		30.5	28.7	64	62.8	100	95.8	89	51.9	67	58.4	70.1	59.5
o4-mini2026.05		24.5	21	37.5	36.6	99.5	74.7	90	51	49.5	19.3	60.2	40.5
GPT-4o-11202026.05		12.5	8.4	4	3.3	47.5	34.5	66	28.5	19.5	7.1	29.9	16.4
Qwen2.5-VL-7B-Instruct2026.05		12.5	4.6	19.5	10.5	66	38.6	50	15	4.5	1.6	30.5	14.1
Qwen2.5-VL-72B-Instruct2026.05		11	4.8	6.5	3.4	95.5	67.9	56	18.7	32	10.8	40.2	21.1
Qwen3-VL-30B-A3B-Instruct2026.05		10	5.4	9.5	5.8	91	65.2	62	27	37	13.8	41.9	23.5
InternVL3.5-241B-A28B2026.05		9	7.9	10.5	9.1	86.5	63.1	68.5	29.4	32.5	10.1	41.4	23.9
Qwen3-VL-30B-A3B-Thinking2026.05		8.5	6.6	2.5	1.7	95	69.3	69.5	33.9	34	13.9	41.9	25.1
Qwen3-VL-8B-Instruct2026.05		7	2.8	2	1	95	68.2	59	26	40	13.9	40.6	22.4
Qwen2.5-VL-32B-Instruct2026.05		5.5	1.7	3	2	75.5	54.8	47.5	16.9	36	12.8	33.5	17.6
Qwen3-VL-235B-A22B-Instruct2026.05		5	4.7	6	4.9	100	76.1	71	36.1	39.5	18.4	44.3	28
InternVL3.5-38B2026.05		3.5	2.9	5	3.7	98	70.1	51.5	18.9	24.5	7.3	36.5	20.6
Qwen3-VL-235B-A22B-Thinking2026.05		3	3.2	2	2	97.5	75.6	82	44.2	46.5	19.2	46.2	28.8
InternVL3.5-8B2026.05		3	1.2	3	2.6	92	66.1	42.5	14.2	32	10.9	34.5	19
Qwen3-VL-32B-Instruct2026.05		2.5	1.1	3	2.1	99	73.2	76.5	39.8	28	12.5	41.8	25.7
Qwen3-VL-32B-Thinking2026.05		1.5	1.1	13.5	12	95.5	69.1	76	39.6	36.5	14.4	44.6	27.2
Qwen3-VL-8B-Thinking2026.05		0.5	0.5	0	0	71.5	50.6	59.5	22.6	32.5	11.2	32.8	17