Interleaved Multi-Image Instruction Following on InterleaveBench
93Image Consistency (2 Obj.)Inset
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Inset#Params.=7B+7B2026.05 | 93 | 94 | 90 | 94 | 93 | 82 | 78 | 72 | 71 | 75 | |
| GPT-4o#Params.=-2026.05 | 92 | 93 | 90 | 89 | 91 | 86 | 85 | 78 | 81 | 81 | |
| Nano Banana#Params.=-2026.05 | 92 | 94 | 83 | 83 | 88 | 83 | 87 | 74 | 76 | 79 | |
| Seedream 4.0#Params.=-2026.05 | 90 | 93 | 91 | 91 | 91 | 84 | 88 | 85 | 85 | 85 | |
| Qwen Image#Params.=7B+20B2026.05 | 85 | 75 | 70 | 41 | 69 | 78 | 66 | 48 | 20 | 49 | |
| DreamOmni2#Params.=7B+5B+12B2026.05 | 83 | 80 | 69 | 65 | 75 | 71 | 67 | 51 | 47 | 57 | |
| FluxKontext#Params.=5B+12B2026.05 | 83 | 80 | 69 | 60 | 74 | 72 | 68 | 50 | 42 | 56 | |
| BAGEL#Params.=7B+7B2026.05 | 68 | 62 | 57 | 57 | 61 | 59 | 48 | 40 | 38 | 45 |