Multi-turn Image Editing on MSE-Bench benchmark leaderboard · SOTA2 Research

Multi-turn Image Editing on MSE-Bench

99.7Success Rate (Turn 1)Nano Banana

Evaluation Results

Method	Links
Nano Banana2025.06		99.7	77.3	75.7	73	64.3
Nano Banana2025.06		98.7	77.3	75.3	72.7	62.7
Qwen-Image-Edit2025.06		98	73.7	66.7	61.3	43
JoyImage-Edit2026.06		98	80.2	67.9	61.7	44.6
Bagel2025.06		96.7	65	61.3	55	41.3
GPT Image 12025.06		96.7	70.7	70	69.7	64
Bagel2025.06		96.3	63	56.7	47.3	30
GPT Image 12025.06		96.3	69	67.3	63.7	55.7
FLUX.1-Kontext (dev)2025.06		95	67	62.3	57.3	44
VINCIE2025.06		95	69.3	66.7	61.7	48.7
VINCIE2026.06		95	69.3	66.7	61.7	48.7
AnchorEdit2026.06		95	73.7	70.7	62.3	56.7
Step1X-Edit2025.06		93.7	54	42	30	14
VINCIE2025.06		91.3	45	39.3	30	21
VINCIE2025.06		91.3	53.3	49.7	44.3	33
GPT-4o2025.06		87.2	78.3	75.5	64.2	49.1
OmniGen2025.06		85.3	18.8	16	12.5	6.5
OmniGen2025.06		84.7	22.3	17	14	8.3
OmniGen22025.06		84.7	39.3	32.7	26.3	13.3
VINCIE2025.06		83.7	51.7	46.3	40	35
MagicBrush2025.06		70.7	30	21.3	17	8.7
UltraEdit2025.06		67.3	23	17.3	11.3	6.7
VINCIE2025.06		66.1	50	32.3	20.9	7
ICEdit2025.06		63.3	34	25.7	16.3	9
Instruct-Pix2Pix2025.06		52	13	11	8.3	6
HQEdit2025.06		47.7	17.7	14	11.3	7.7
OmniGen2025.06		33.3	3.5	0.2	0	0
UltraEdit2025.06		31	6.2	1.5	0.2	0
HQEdit2025.06		17	7.3	2	0.3	0