Long-Horizon World Modelling on WorldPredictionBench v1 (test)
4.31COINBAGEL
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| BAGELEvaluator=GPT-4o-as-a-judge2026.02 | 4.31 | 3.94 | 2.88 | 3.37 | 3.01 | 3.3 | 4.29 | 4.1 | 3.47 | 3.22 | 2.47 | 2.23 | |
| OMNIGEN2Evaluator=GPT-4o-as-a-judge2026.02 | 3.78 | 3.53 | 2.73 | 3.01 | 3.24 | 3.28 | 3.25 | 2.64 | 4.05 | 3.75 | 3.06 | 3.11 | |
| OMNIGENEvaluator=GPT-4o-as-a-judge2026.02 | 3.61 | 3.48 | 3.2 | 3.14 | 3.56 | 3.45 | 3.25 | 2.83 | 4 | 3.92 | 3.28 | 3.45 | |
| UNIWORLD-V1Evaluator=GPT-4o-as-a-judge2026.02 | 3.27 | 3.42 | 3.2 | 2.57 | 3.39 | 3.35 | 3.4 | 3.41 | 3.68 | 3.59 | 3.09 | 2.97 | |
| SWIRL (IDM → FWM)Evaluator=GPT-4o-as-a-judge, Strategy=IDM → FWM2026.02 | 2.61 | 2.45 | 1.84 | 1.95 | 1.56 | 1.89 | 3.24 | 2.42 | 1.85 | 1.59 | 1.25 | 1.08 | |
| SWIRL (ITERATIVE)Evaluator=GPT-4o-as-a-judge, Strategy=Iterative2026.02 | 2.22 | 2.01 | 1.76 | 1.86 | 1.49 | 1.74 | 3.23 | 2.08 | 1.53 | 1.32 | 1.15 | 1.11 | |
| LIQUID-SFTEvaluator=GPT-4o-as-a-judge2026.02 | 2.11 | 2.06 | 1.59 | 1.8 | 1.35 | 1.63 | 3.09 | 2.09 | 1.4 | 1.17 | 1.07 | 0.97 | |
| SWIRL (ITER.+SHARE)Evaluator=GPT-4o-as-a-judge, Strategy=Iterative + Share2026.02 | 2.06 | 2.28 | 1.82 | 1.66 | 1.43 | 1.68 | 2.95 | 2.04 | 1.53 | 1.24 | 1.23 | 1.11 | |
| BLIP3O-NEXTEvaluator=GPT-4o-as-a-judge2026.02 | 1.74 | 1.58 | 1.24 | 1.37 | 1.7 | 1.31 | 2.46 | 1.82 | 1.2 | 0.95 | 0.83 | 0.63 |