Continual Learning on MLLM-CL Ability benchmark leaderboard · SOTA2 Research

Continual Learning on MLLM-CL Ability

38.4OCR Scoreϕ-DPO

Evaluation Results

Method	Links
ϕ-DPO2026.02		38.4	39.2	68.65	35	45.55	45.31	43.03	-0.31
DISCO*2026.02		34.2	35	61.55	27.5	40.14	39.56	37.85	-0.77
MR-LoRA2026.02		33.7	36.2	65.1	32.5	41.89	41.88	38.86	-0.02
DISCO2026.02		32.9	33.1	60.15	30.14	39.02	39.07	36.57	0.07
Zeroshot2026.02		31.2	30.2	60.79	10	33.05	—	—	—
O-LoRA*2026.02		29.6	31.3	60.79	27.5	39.96	37.3	36.34	-3.55
O-LoRA2026.02		29.6	32.9	52.41	33.75	39.72	37.16	35.42	-3.41
MoELoRA2026.02		26.7	32.8	56.85	27.22	39.45	35.89	36.07	-4.75
SEFE2026.02		26	33.4	57.74	33.75	40.98	37.72	36.59	-4.35
SEFE*2026.02		25.6	34.8	57.61	31.39	42.25	37.35	37.93	-6.53
CL-MoE*2026.02		25.4	31.8	60.91	30	41.22	37.03	37.28	-5.59
HiDe*2026.02		24.6	28.4	30.71	23.75	36.84	26.86	33.54	-13.3
HiDe2026.02		24.6	32.1	46.32	28.75	37.98	32.94	34.6	-6.72
LoRA-FT2026.02		23.6	33.7	55.84	32.5	41.28	36.41	36.58	-6.49
LoRA-FT*2026.02		21.8	32.7	58.38	28.75	40.32	35.41	36.32	-6.55
CL-MoE2026.02		19.9	32.7	53.43	30.69	40.5	34.18	35.65	-8.43
MoELoRA*2026.02		19.8	32.2	54.19	30	40.35	34.05	35.39	-8.41