Routing on SPROUT (test)
98.6AccuracyOracle
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Oracle2026.01 | 98.6 | 0.6 | — | 92.2 | |
| PROTEUSEncoder=DeBERTa-v3-small, Encoder parameters=22M, Batch size=82026.01 | 94 | 0.93 | 9.5 | 83.5 | |
| Best Fixed (o3-mini)2026.01 | 90.7 | 5.5 | — | 25.8 | |
| CARROT-QualityEncoder parameters=22–125M, Batch size=82026.01 | 89.9 | 2.24 | 8 | 60.9 | |
| CARROT-BalancedEncoder parameters=22–125M, Batch size=82026.01 | 83.9 | 0.35 | 6.8 | 81 | |
| OmniRouterEncoder parameters=22–125M, Batch size=82026.01 | 82.6 | 0.64 | 6 | 76.6 | |
| MLPEncoder parameters=22–125M, Batch size=82026.01 | 77.1 | 1.7 | 2.3 | 60.7 | |
| CARROT-CostEncoder parameters=22–125M, Batch size=82026.01 | 74 | 0.11 | 1.2 | 69.9 | |
| KNNEncoder parameters=22–125M, Batch size=82026.01 | 73.8 | 1.7 | 0.9 | 58.1 | |
| Random2026.01 | 71.8 | 1.7 | — | 56.8 | |
| Cheapest2026.01 | 63.5 | 0.06 | — | 63.8 |