Task Success Metrics (Real, WM, W2R, CR) on ALFWorld
91Real SuccessGPT-5
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-5World Model=Qwen2.5-7B WorldModel2025.12 | 91 | 84.62 | 86.67 | 0.95 | |
| GPT-5World Model=Llama3.1-8B WorldModel2025.12 | 91 | 82.56 | 81.54 | 0.9 | |
| Claude-sonnet-4.5World Model=Qwen2.5-7B WorldModel2025.12 | 82 | 76 | 76 | 0.93 | |
| Claude-sonnet-4.5World Model=Llama3.1-8B WorldModel2025.12 | 82 | 84 | 84 | 1.02 | |
| GPT-4-turboWorld Model=Qwen2.5-7B WorldModel2025.12 | 74.21 | 62.56 | 64.62 | 0.87 | |
| GPT-4-turboWorld Model=Llama3.1-8B WorldModel2025.12 | 74.21 | 67.53 | 67.01 | 0.9 | |
| GPT-4.1World Model=Qwen2.5-7B WorldModel2025.12 | 67.2 | 68.56 | 69.59 | 1.04 | |
| GPT-4.1World Model=Llama3.1-8B WorldModel2025.12 | 67.2 | 68.72 | 68.21 | 1.02 | |
| GPT-4oWorld Model=Qwen2.5-7B WorldModel2025.12 | 58 | 55.9 | 57.44 | 0.99 | |
| GPT-4oWorld Model=Llama3.1-8B WorldModel2025.12 | 58 | 58.46 | 56.92 | 0.98 | |
| Gemini-2.5-flashWorld Model=Qwen2.5-7B WorldModel2025.12 | 50.5 | 51.79 | 52.31 | 1.04 | |
| Gemini-2.5-flashWorld Model=Llama3.1-8B WorldModel2025.12 | 50.5 | 53.33 | 53.33 | 1.06 | |
| GPT-4o-miniWorld Model=Qwen2.5-7B WorldModel2025.12 | 7.69 | 7.69 | 7.69 | 1 | |
| GPT-4o-miniWorld Model=Llama3.1-8B WorldModel2025.12 | 7.69 | 9.74 | 9.74 | 1.27 |