Query Routing on LongBench OOD v2
53QAAlways-LC (Baseline)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Always-LC (Baseline)Answer Model=DeepSeek-R1 [T]2026.05 | 53 | 100 | 49 | |
| Always-LC (Baseline)Answer Model=Qwen3-235B [T]2026.05 | 52 | 100 | 54 | |
| Pre-Route (Q235B) [N]Answer Model=DeepSeek-R1 [T]2026.05 | 51 | 41.8 | 56 | |
| Pre-Route (Q235B) [T]Answer Model=DeepSeek-R1 [T]2026.05 | 51 | 31.1 | 65 | |
| Self-Route (Baseline)Answer Model=Qwen3-235B [T]2026.05 | 49 | 52.2 | 52 | |
| Pre-Route (D-Q1.7B) [T]Answer Model=DeepSeek-R1 [T]2026.05 | 48 | 21.9 | 66 | |
| Always-LC (Baseline)Answer Model=Qwen-Max [N]2026.05 | 48 | 100 | 59 | |
| Pre-Route (R1) [T]Answer Model=Qwen3-235B [N]2026.05 | 47 | 23.5 | 72 | |
| Pre-Route (Q235B) [N]Answer Model=Qwen3-235B [N]2026.05 | 47 | 37.4 | 62 | |
| Pre-Route (Q1.7B) [N]Answer Model=Qwen3-235B [N]2026.05 | 47 | 36.7 | 61 | |
| Pre-Route (D-Q1.7B) [N]Answer Model=Qwen3-235B [N]2026.05 | 47 | 27.9 | 67 | |
| Pre-Route (Q235B) [N]Answer Model=Qwen3-235B [T]2026.05 | 47 | 41.3 | 55 | |
| Pre-Route (Q235B) [T]Answer Model=Qwen3-235B [T]2026.05 | 47 | 28.5 | 61 | |
| Pre-Route (Q1.7B) [N]Answer Model=Qwen3-235B [T]2026.05 | 47 | 37.1 | 55 | |
| Pre-Route (D-Q1.7B) [N]Answer Model=Qwen3-235B [T]2026.05 | 47 | 28.7 | 61 | |
| Pre-Route (R1) [T]Answer Model=DeepSeek-R1 [T]2026.05 | 47 | 27.6 | 61 | |
| Pre-Route (Q1.7B) [N]Answer Model=DeepSeek-R1 [T]2026.05 | 47 | 42.3 | 53 | |
| Pre-Route (Q1.7B) [T]Answer Model=DeepSeek-R1 [T]2026.05 | 47 | 41.3 | 57 | |
| Always-LC (Baseline)Answer Model=Qwen3-235B [N]2026.05 | 46 | 100 | 60 | |
| Pre-Route (Q1.7B) [T]Answer Model=Qwen3-235B [N]2026.05 | 46 | 40.5 | 56 | |
| Pre-Route (D-Q1.7B) [T]Answer Model=Qwen3-235B [N]2026.05 | 46 | 26.7 | 66 | |
| Pre-Route (Q1.7B) [T]Answer Model=Qwen3-235B [T]2026.05 | 46 | 43 | 55 | |
| Pre-Route (D-Q1.7B) [T]Answer Model=Qwen3-235B [T]2026.05 | 46 | 29 | 58 | |
| Self-Route (Baseline)Answer Model=DeepSeek-R1 [T]2026.05 | 46 | 44.4 | 53 | |
| Pre-Route (D-Q1.7B) [N]Answer Model=DeepSeek-R1 [T]2026.05 | 46 | 18.4 | 64 | |
| Self-Route (Baseline)Answer Model=Qwen3-235B [N]2026.05 | 45 | 60.9 | 44 | |
| Pre-Route (Q235B) [T]Answer Model=Qwen3-235B [N]2026.05 | 45 | 27.4 | 67 | |
| Pre-Route (R1) [T]Answer Model=Qwen3-235B [T]2026.05 | 45 | 23.8 | 63 | |
| Always-RAG (Baseline)Answer Model=DeepSeek-R1 [T]2026.05 | 45 | 0 | 51 | |
| Self-Route (Baseline)Answer Model=Qwen-Max [N]2026.05 | 45 | 57.4 | 46 | |
| Pre-Route (Q1.7B) [N]Answer Model=Qwen-Max [N]2026.05 | 45 | 38.7 | 60 | |
| Pre-Route (D-Q1.7B) [T]Answer Model=Qwen-Max [N]2026.05 | 45 | 27.7 | 66 | |
| Pre-Route (Q235B) [N]Answer Model=Qwen-Max [N]2026.05 | 44 | 40 | 59 | |
| Pre-Route (Q235B) [T]Answer Model=Qwen-Max [N]2026.05 | 44 | 29 | 65 | |
| Pre-Route (Q1.7B) [T]Answer Model=Qwen-Max [N]2026.05 | 44 | 41.8 | 56 | |
| Pre-Route (D-Q1.7B) [N]Answer Model=Qwen-Max [N]2026.05 | 44 | 27.3 | 66 | |
| Always-RAG (Baseline)Answer Model=Qwen3-235B [T]2026.05 | 43 | 0 | 46 | |
| Pre-Route (R1) [T]Answer Model=Qwen-Max [N]2026.05 | 43 | 25.8 | 68 | |
| Pre-Route (Q235B) [N]Answer Model=Qwen3-30B [N]2026.05 | 42 | 39 | 60 | |
| Pre-Route (D-Q1.7B) [N]Answer Model=Qwen3-30B [N]2026.05 | 42 | 27.6 | 68 | |
| Pre-Route (D-Q1.7B) [T]Answer Model=Qwen3-30B [N]2026.05 | 42 | 28 | 67 | |
| Always-RAG (Baseline)Answer Model=Qwen3-235B [N]2026.05 | 42 | 0 | 40 | |
| Self-Route (Baseline)Answer Model=Qwen3-8B [T]2026.05 | 41 | 32.3 | 67 | |
| Pre-Route (R1) [T]Answer Model=Qwen3-30B [N]2026.05 | 41 | 25.9 | 70 | |
| Pre-Route (Q235B) [T]Answer Model=Qwen3-30B [N]2026.05 | 41 | 29.2 | 67 | |
| Always-LC (Baseline)Answer Model=Qwen3-8B [N]2026.05 | 40 | 100 | 40 | |
| Always-LC (Baseline)Answer Model=Qwen3-8B [T]2026.05 | 40 | 100 | 37 | |
| Pre-Route (Q235B) [N]Answer Model=Qwen3-8B [T]2026.05 | 40 | 33 | 63 | |
| Always-LC (Baseline)Answer Model=Qwen3-30B [N]2026.05 | 40 | 100 | 40 | |
| Self-Route (Baseline)Answer Model=Qwen3-30B [N]2026.05 | 40 | 38 | 61 | |
| Pre-Route (Q1.7B) [T]Answer Model=Qwen3-30B [N]2026.05 | 40 | 41 | 57 | |
| Always-RAG (Baseline)Answer Model=Qwen-Max [N]2026.05 | 40 | 0 | 41 | |
| Pre-Route (R1) [T]Answer Model=Qwen3-8B [T]2026.05 | 39 | 19.9 | 71 | |
| Always-RAG (Baseline)Answer Model=Qwen3-30B [N]2026.05 | 39 | 0 | 60 | |
| Self-Route (Baseline)Answer Model=Qwen3-8B [N]2026.05 | 38 | 31.3 | 64 | |
| Pre-Route (Q235B) [N]Answer Model=Qwen3-8B [N]2026.05 | 38 | 31.8 | 64 | |
| Pre-Route (Q1.7B) [T]Answer Model=Qwen3-8B [N]2026.05 | 38 | 38.9 | 61 | |
| Pre-Route (D-Q1.7B) [N]Answer Model=Qwen3-8B [N]2026.05 | 38 | 20.7 | 70 | |
| Pre-Route (Q235B) [T]Answer Model=Qwen3-8B [T]2026.05 | 38 | 29.6 | 63 | |
| Pre-Route (Q1.7B) [N]Answer Model=Qwen3-8B [T]2026.05 | 38 | 38.1 | 58 | |
| Pre-Route (D-Q1.7B) [N]Answer Model=Qwen3-8B [T]2026.05 | 38 | 21.9 | 68 | |
| Pre-Route (Q1.7B) [N]Answer Model=Qwen3-30B [N]2026.05 | 38 | 39.9 | 57 | |
| Always-LC (Baseline)Answer Model=Qwen3-4B [T]2026.05 | 37 | 100 | 40 | |
| Self-Route (Baseline)Answer Model=Qwen3-4B [T]2026.05 | 37 | 33.8 | 65 | |
| Pre-Route (R1) [T]Answer Model=Qwen3-8B [N]2026.05 | 37 | 19.6 | 72 | |
| Pre-Route (Q235B) [T]Answer Model=Qwen3-8B [N]2026.05 | 37 | 25.8 | 68 | |
| Always-RAG (Baseline)Answer Model=Qwen3-8B [T]2026.05 | 37 | 0 | 63 | |
| Pre-Route (Q1.7B) [T]Answer Model=Qwen3-8B [T]2026.05 | 37 | 40 | 54 | |
| Pre-Route (Q235B) [N]Answer Model=Qwen3-4B [T]2026.05 | 36 | 38.7 | 59 | |
| Pre-Route (D-Q1.7B) [T]Answer Model=Qwen3-8B [N]2026.05 | 36 | 18.9 | 69 | |
| Pre-Route (D-Q1.7B) [T]Answer Model=Qwen3-8B [T]2026.05 | 36 | 21.2 | 68 | |
| Always-LC (Baseline)Answer Model=Qwen3-30B [T]2026.05 | 36 | 100 | 56 | |
| Pre-Route (R1) [T]Answer Model=Qwen3-4B [T]2026.05 | 35 | 24.5 | 70 | |
| Pre-Route (Q1.7B) [N]Answer Model=Qwen3-8B [N]2026.05 | 35 | 39.8 | 55 | |
| Always-RAG (Baseline)Answer Model=Qwen3-1.7B [T]2026.05 | 34 | 0 | 69 | |
| Always-RAG (Baseline)Answer Model=Qwen3-4B [T]2026.05 | 34 | 0 | 60 | |
| Pre-Route (Q235B) [T]Answer Model=Qwen3-4B [T]2026.05 | 34 | 30.1 | 64 | |
| Pre-Route (Q1.7B) [N]Answer Model=Qwen3-4B [T]2026.05 | 34 | 38.2 | 58 | |
| Pre-Route (Q1.7B) [T]Answer Model=Qwen3-4B [T]2026.05 | 34 | 40.7 | 55 | |
| Pre-Route (D-Q1.7B) [N]Answer Model=Qwen3-4B [T]2026.05 | 34 | 21.1 | 67 | |
| Pre-Route (D-Q1.7B) [T]Answer Model=Qwen3-4B [T]2026.05 | 34 | 20.4 | 66 | |
| Always-RAG (Baseline)Answer Model=Qwen3-8B [N]2026.05 | 34 | 0 | 60 | |
| Always-RAG (Baseline)Answer Model=Qwen3-1.7B [N]2026.05 | 33 | 0 | 62 | |
| Self-Route (Baseline)Answer Model=Qwen3-1.7B [T]2026.05 | 33 | 28.1 | 68 | |
| Pre-Route (R1) [T]Answer Model=Qwen3-1.7B [T]2026.05 | 33 | 23.5 | 71 | |
| Pre-Route (D-Q1.7B) [N]Answer Model=Qwen3-1.7B [T]2026.05 | 33 | 8 | 80 | |
| Pre-Route (D-Q1.7B) [T]Answer Model=Qwen3-1.7B [T]2026.05 | 33 | 7.5 | 80 | |
| Self-Route (Baseline)Answer Model=Qwen3-1.7B [N]2026.05 | 32 | 34.9 | 63 | |
| Pre-Route (D-Q1.7B) [T]Answer Model=Qwen3-1.7B [N]2026.05 | 32 | 6.6 | 82 | |
| Pre-Route (Q235B) [T]Answer Model=Qwen3-1.7B [T]2026.05 | 32 | 29.4 | 65 | |
| Pre-Route (Q1.7B) [T]Answer Model=Qwen3-1.7B [T]2026.05 | 32 | 38.7 | 58 | |
| Pre-Route (Q235B) [T]Answer Model=Qwen3-4B [N]2026.05 | 32 | 27 | 69 | |
| Pre-Route (Q1.7B) [T]Answer Model=Qwen3-4B [N]2026.05 | 32 | 42.1 | 58 | |
| Pre-Route (Q235B) [N]Answer Model=Qwen3-1.7B [N]2026.05 | 31 | 39.7 | 58 | |
| Pre-Route (Q235B) [T]Answer Model=Qwen3-1.7B [N]2026.05 | 31 | 29.7 | 66 | |
| Pre-Route (D-Q1.7B) [N]Answer Model=Qwen3-1.7B [N]2026.05 | 31 | 7.3 | 81 | |
| Pre-Route (Q235B) [N]Answer Model=Qwen3-1.7B [T]2026.05 | 31 | 37.8 | 60 | |
| Pre-Route (Q1.7B) [N]Answer Model=Qwen3-1.7B [T]2026.05 | 31 | 38.9 | 56 | |
| Always-RAG (Baseline)Answer Model=Qwen3-4B [N]2026.05 | 31 | 0 | 70 | |
| Pre-Route (R1) [T]Answer Model=Qwen3-4B [N]2026.05 | 31 | 24.7 | 69 |