Multi-judge evaluation on Shared 500-prompt sample
0.87Global Correlation (r)GPT-5.2
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-5.2Family=OpenAI2026.03 | 0.87 | 0.7 | 20 | 69.4 | |
| Claude Sonnet 4Family=Anthropic2026.03 | 0.59 | 0.42 | 29 | 47.7 | |
| GPT-4.1-miniFamily=OpenAI2026.03 | 0.56 | 0.47 | 17 | 43.6 | |
| Gemini-2.5-flashFamily=Google2026.03 | 0.47 | 0.27 | 42 | 23.8 | |
| Llama-3.3-70bFamily=Meta2026.03 | 0.31 | 0.23 | 25 | 18.6 |