Future Event Prediction on FutureBench (test)
80.91-Hop ScoreQwen2.5-VL-7B CoE-GRPO
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Qwen2.5-VL-7B CoE-GRPOMethod=CoE-GRPO, Frames=322026.03 | 80.9 | 83.9 | 71.6 | 71.4 | 75 | |
| Qwen2.5-VL-3B CoE-GRPOMethod=CoE-GRPO, Frames=322026.03 | 71.1 | 73.6 | 69.7 | 64.6 | 68.28 | |
| Qwen2.5-VL-3B CoE-SFTMethod=CoE-SFT, Frames=322026.03 | 68.8 | 75.1 | 55.7 | 60.5 | 63.6 | |
| Qwen2.5-VL-7B CoE-SFTMethod=CoE-SFT, Frames=322026.03 | 67.6 | 74.1 | 62.2 | 63.2 | 65.72 | |
| Qwen2.5-VL-32BMethod=Vanilla, Frames=322026.03 | 66.5 | 62.7 | 63.2 | 55.2 | 59.94 | |
| Qwen2.5-VL-7B (NEP) GRPOMethod=(NEP) GRPO, Frames=322026.03 | 66.2 | 69.9 | 63.7 | 68.1 | 67.28 | |
| Qwen3-VL-30B-A3BMethod=Vanilla, Frames=322026.03 | 65.3 | 70.5 | 76.1 | 62.2 | 66.86 | |
| Qwen2.5-VL-3B (NEP) GRPOMethod=(NEP) GRPO, Frames=322026.03 | 63 | 65.3 | 63.7 | 67.1 | 65.45 | |
| GPT-4oMethod=Vanilla, Frames=322026.03 | 61.9 | 61.7 | 72.1 | 51.6 | 59.04 | |
| Qwen2.5-VL-7B (NEP) SFTMethod=(NEP) SFT, Frames=322026.03 | 60.7 | 74.2 | 66.2 | 65 | 64.39 | |
| GPT-5Method=Vanilla, Frames=322026.03 | 59.6 | 57.3 | 62.6 | 55.6 | 57.92 | |
| MiMo-VL-7BMethod=Vanilla, Frames=322026.03 | 59 | 59.6 | 50.5 | 43.8 | 50.45 | |
| Qwen2.5-VL-7B InstructMethod=Instruct, Frames=322026.03 | 57.2 | 57 | 50.2 | 50.7 | 52.94 | |
| Qwen2.5-VL-72BMethod=Vanilla, Frames=322026.03 | 55.5 | 68.4 | 63.7 | 53.2 | 58.33 | |
| InternVL3-8BMethod=Vanilla, Frames=322026.03 | 54.3 | 58 | 63.2 | 54.4 | 56.72 | |
| Qwen2.5-VL-3B (NEP) SFTMethod=(NEP) SFT, Frames=322026.03 | 53.2 | 57 | 59.2 | 59.3 | 57.86 | |
| LLavA-NeXT-VideoMethod=Vanilla, Frames=322026.03 | 48.8 | 49.3 | 40 | 44.4 | 45.24 | |
| Qwen2.5-VL-3B InstructMethod=Instruct, Frames=322026.03 | 45.1 | 50.8 | 44.3 | 45.8 | 46.3 | |
| Kimi-VL-A3BMethod=Vanilla, Frames=322026.03 | 44.3 | 42.8 | 51.3 | 51.9 | 48.87 | |
| GLM-4.1V-9BMethod=Vanilla, Frames=322026.03 | 29.9 | 41.9 | 52.2 | 47.3 | 44.38 |