Causal Variable Identification on CRASS
92.3F1 (X)GPT-5
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-5Setting=Decompositional Framework2025.05 | 92.3 | 91.1 | 87.4 | 91.7 | |
| GPT-o4Setting=Decompositional Framework, Model Variant=GPT-o4-mini-high2025.05 | 91 | 89.2 | 84.1 | 89.3 | |
| Llama4-MSetting=Decompositional Framework, Model Variant=Llama-4-Maverick-17B-128E2025.05 | 90.6 | 89.1 | 82.4 | 88.6 | |
| DeepSeekSetting=Decompositional Framework, Model Variant=DeepSeek-VL2025.05 | 89.5 | 87.1 | 76.2 | 83.5 | |
| Gemini2.5Setting=Decompositional Framework, Model Variant=Gemini2.5-Pro2025.05 | 88.6 | 86.2 | 74.1 | 81.6 | |
| Llama4-SSetting=Decompositional Framework, Model Variant=Llama-4-Scout-17B2025.05 | 88.5 | 86.9 | 81.2 | 87.4 | |
| Qwen3Setting=Decompositional Framework, Model Variant=Qwen3-VL-235B-A22B Thinking2025.05 | 87.3 | 85.4 | 72.8 | 79.9 |