Reranking on human pairwise T10 (test)
78EM (%)Kimi-K2.6
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Kimi-K2.6Inference path=Direct judge2026.07 | 78 | — | |
| SimPOInference path=log p(Yes)2026.07 | 78 | 4.28 | |
| BaseInference path=Direct judge2026.07 | 75.6 | 0.76 | |
| SFTInference path=Direct judge2026.07 | 75.6 | — | |
| BaseInference path=log p(Yes)2026.07 | 63.4 | 0.76 | |
| RandomInference path=Direct judge2026.07 | 50 | — | |
| RandomInference path=log p(Yes)2026.07 | 50 | — |