Test-Time Scaling Selection on RouterBench Quality-Priority
0.7924RewardOracle
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| OracleNumber of seeds=5, Iterations=160, Warm-up phase=50-step2026.05 | 0.7924 | 68.12 | 115.7 | |
| OracleBackbone=Qwen3-0.6B, Number of seeds=5, Iterations=160, Warm-up phase=50-step2026.05 | 0.7426 | 59.12 | 20.3 | |
| UNISCALEBackbone=Qwen3-0.6B, Number of seeds=5, Iterations=160, Warm-up phase=50-step2026.05 | 0.645 | 48.12 | 26.6 | |
| UNISCALENumber of seeds=5, Iterations=160, Warm-up phase=50-step2026.05 | 0.6306 | 57.37 | 1,374.7 | |
| TSNumber of seeds=5, Iterations=160, Warm-up phase=50-step2026.05 | 0.6243 | 54 | 1,875.7 | |
| GreedyBackbone=Qwen3-0.6B, Number of seeds=5, Iterations=160, Warm-up phase=50-step2026.05 | 0.6184 | 46.5 | 67.6 | |
| RandomNumber of seeds=5, Iterations=160, Warm-up phase=50-step2026.05 | 0.6175 | 52.88 | 1,359.8 | |
| MLPNumber of seeds=5, Iterations=160, Warm-up phase=50-step2026.05 | 0.6055 | 47.5 | 353.4 | |
| NeuralUCBNumber of seeds=5, Iterations=160, Warm-up phase=50-step2026.05 | 0.5929 | 50.25 | 1,819.7 | |
| TSBackbone=Qwen3-0.6B, Number of seeds=5, Iterations=160, Warm-up phase=50-step2026.05 | 0.5901 | 44.88 | 63.9 | |
| k-NNNumber of seeds=5, Iterations=160, Warm-up phase=50-step2026.05 | 0.5807 | 46.75 | 1,113.4 | |
| GreedyNumber of seeds=5, Iterations=160, Warm-up phase=50-step2026.05 | 0.578 | 52.88 | 3,402.1 | |
| MLPBackbone=Qwen3-0.6B, Number of seeds=5, Iterations=160, Warm-up phase=50-step2026.05 | 0.5752 | 42.38 | 46.1 | |
| RandomBackbone=Qwen3-0.6B, Number of seeds=5, Iterations=160, Warm-up phase=50-step2026.05 | 0.5726 | 42.25 | 73.2 | |
| NeuralUCBBackbone=Qwen3-0.6B, Number of seeds=5, Iterations=160, Warm-up phase=50-step2026.05 | 0.558 | 41.12 | 68.4 | |
| k-NNBackbone=Qwen3-0.6B, Number of seeds=5, Iterations=160, Warm-up phase=50-step2026.05 | 0.5146 | 36.25 | 47.8 |