Next Query Prediction on NQP-Bench Average
47.9Judge ScoreOnePred
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| OnePredModel Regime=Gemini-3.1-Pro, History Context Strategy=Ours2026.05 | 47.9 | 48.55 | |
| Gemini-3.1-ProModel Regime=Gemini-3.1-Pro, History Context Strategy=Full-history2026.05 | 45.31 | 45.71 | |
| OnePredModel Regime=RL-trained Qwen, History Context Strategy=Ours2026.05 | 44 | 44.98 | |
| Gemini-3.1-ProModel Regime=Gemini-3.1-Pro, History Context Strategy=Current-turn2026.05 | 43.16 | 43.58 | |
| RL-trained QwenModel Regime=RL-trained Qwen, History Context Strategy=Full-history2026.05 | 42.43 | 42.71 | |
| RL-trained QwenModel Regime=RL-trained Qwen, History Context Strategy=Current-turn2026.05 | 40.9 | 41.02 | |
| OnePredModel Regime=Base Qwen, History Context Strategy=Ours2026.05 | 36.96 | 37.07 | |
| Base QwenModel Regime=Base Qwen, History Context Strategy=Full-history2026.05 | 35.09 | 35.77 | |
| Base QwenModel Regime=Base Qwen, History Context Strategy=Current-turn2026.05 | 33.34 | 34.12 |