World Modeling on WorldPrediction-WM
65.7AccuracyVL-JEPA SFT
Evaluation Results
| Method | Links | |
|---|---|---|
| VL-JEPA SFTParameter Scale=1.6B, Model Variant=SFT2025.12 | 65.7 | |
| VL-JEPA BASEParameter Scale=1.6B, Model Variant=BASE2025.12 | 63.9 | |
| Qwen2.5-VLModel Category=Vision Language Models, Parameter Scale=72B2025.12 | 57 | |
| Gemini-2Model Category=Socratic LLMs, Captioning Model=Qwen2.5-VL-72B2025.12 | 55.6 | |
| Llama-4Model Category=Socratic LLMs, Parameter Scale=400B, Captioning Model=Qwen2.5-VL-72B2025.12 | 53.6 | |
| Claude-3.5Model Category=Socratic LLMs, Captioning Model=Qwen2.5-VL-72B2025.12 | 53.3 | |
| Llama-4Model Category=Socratic LLMs, Parameter Scale=109B, Captioning Model=Qwen2.5-VL-72B2025.12 | 52.7 | |
| GPT-4oModel Category=Socratic LLMs, Captioning Model=Qwen2.5-VL-72B2025.12 | 52 | |
| InternVL2.5Model Category=Vision Language Models, Parameter Scale=38B2025.12 | 50.3 | |
| Llama-3.1Model Category=Socratic LLMs, Parameter Scale=70B, Captioning Model=Qwen2.5-VL-72B2025.12 | 49.8 | |
| Qwen2.5Model Category=Socratic LLMs, Parameter Scale=7B, Captioning Model=Qwen2.5-VL-72B2025.12 | 49.1 | |
| Qwen2.5-VLModel Category=Vision Language Models, Parameter Scale=32B2025.12 | 49 | |
| Llama-3.1Model Category=Socratic LLMs, Parameter Scale=8B, Captioning Model=Qwen2.5-VL-72B2025.12 | 48.7 | |
| Qwen2.5Model Category=Socratic LLMs, Parameter Scale=72B, Captioning Model=Qwen2.5-VL-72B2025.12 | 48.5 | |
| Qwen2.5-VLModel Category=Vision Language Models, Parameter Scale=7B2025.12 | 45.5 | |
| Qwen2.5Model Category=Socratic LLMs, Parameter Scale=3B, Captioning Model=Qwen2.5-VL-72B2025.12 | 44 | |
| InternVL2.5Model Category=Vision Language Models, Parameter Scale=26B2025.12 | 30.2 | |
| InternVL2.5Model Category=Vision Language Models, Parameter Scale=4B2025.12 | 29.8 | |
| Qwen2.5-VLModel Category=Vision Language Models, Parameter Scale=3B2025.12 | 21.6 | |
| InternVL2.5Model Category=Vision Language Models, Parameter Scale=2B2025.12 | 20 |