End-to-end Retrieval and Reasoning on Set 2 (n=31 queries, top-10 labeled)
91.9Precision@5Kimi-K2.6 (think)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Kimi-K2.6 (think)Configuration=→ Kimi-K2.6 (think), Upstream=SimPO text reranker2026.07 | 91.9 | 77.6 | 85 | |
| Kimi-K2.6 (no think)Configuration=→ Kimi-K2.6 (no think), Upstream=SimPO text reranker2026.07 | 91.1 | 75.5 | 83.5 | |
| BaseConfiguration=→ Base, Upstream=SimPO text reranker2026.07 | 89.3 | 73 | 80.9 | |
| SFT+GRPOConfiguration=→ SFT+GRPO, Upstream=SimPO text reranker2026.07 | 87.6 | 69.5 | 78.3 | |
| SimPO rerank (text)Configuration=+ SimPO rerank (text)2026.07 | 80 | 60.3 | 71.3 | |
| Retrieval only @5Configuration=Retrieval only @52026.07 | 73.3 | 53.8 | 64.8 | |
| RandomConfiguration=Random2026.07 | 68.2 | 50.2 | 61 |