Quality-judgment accuracy on LLMBar ZH condition
86.5Strict Accuracy (%)DeepSeek
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| DeepSeekLanguage condition=ZH, N=8382026.06 | 86.5 | 87.8 | 85.6 | 53.1 | 2.5 | |
| Gemini FlashLanguage condition=ZH, N=8382026.06 | 83.3 | 83.7 | 81.2 | 56.9 | 0.8 | |
| Claude HaikuLanguage condition=ZH, N=8382026.06 | 74 | 74.3 | 71.4 | 53 | 0.7 | |
| GPT-4.1 MiniLanguage condition=ZH, N=8382026.06 | 71.7 | 72.8 | 69.8 | 52.6 | 2.1 |