Human-Model Agreement on AIReg-Bench
0.863Quadratic Weighted KappaGemini 2.5 Pro
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Gemini 2.5 Pro2026.02 | 0.863 | 0.856 | -0.225 | 0.458 | |
| GPT-52026.02 | 0.849 | 0.838 | -0.067 | 0.45 | |
| Grok 42026.02 | 0.829 | 0.829 | 0.242 | 0.475 | |
| GPT-4o2026.02 | 0.775 | 0.842 | 0.458 | 0.558 | |
| Claude Sonnet 42026.02 | 0.772 | 0.779 | -0.15 | 0.6 | |
| Grok 3 mini2026.02 | 0.73 | 0.81 | 0.492 | 0.592 | |
| Gemini 2.5 Flash2026.02 | 0.729 | 0.825 | -0.108 | 0.625 | |
| o32026.02 | 0.723 | 0.809 | -0.192 | 0.658 | |
| Gemma 32026.02 | 0.696 | 0.757 | 0.258 | 0.692 | |
| o3 mini2026.02 | 0.624 | 0.799 | 0.742 | 0.785 |