Agentic Model Routing on OOD n=176 (test)
75.89Average Performance (%)Oracle
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| OracleTaxonomy=Oracle2026.06 | 75.89 | 0 | 2.32 | |
| ACRouterTaxonomy=Agent-as-a-Router2026.06 | 62.5 | 17 | 1.18 | |
| Always-Opus 4.6Taxonomy=Single-Model Baselines2026.06 | 57.14 | 26.7 | 0.64 | |
| Qwen3.5-0.8B-FinetunedTaxonomy=Static: Trained Policy2026.06 | 55.36 | 27.2 | 0.74 | |
| LinUCBTaxonomy=Dynamic: Online Bandit2026.06 | 49.82 | 31.1 | 0.96 | |
| LinTSTaxonomy=Dynamic: Online Bandit2026.06 | 46.43 | 35.9 | 0.75 | |
| RandomTaxonomy=Single-Model Baselines2026.06 | 31.25 | 50.4 | 0.85 | |
| RouteLLM-BERTTaxonomy=Static: Trained Policy2026.06 | 21.43 | 59.4 | 1.3 | |
| LogRegTaxonomy=Static: Trained Policy2026.06 | 19.64 | 61.8 | 1.17 | |
| Always-Kimi-K2.5Taxonomy=Single-Model Baselines2026.06 | 18.75 | 62.3 | 1.22 | |
| kNN RetrievalTaxonomy=Static: Heuristic2026.06 | 14.29 | 66.7 | 1.45 | |
| TF-IDF+MLPTaxonomy=Static: Trained Policy2026.06 | 13.39 | 67.9 | 1.17 | |
| RouteLLM-MFTaxonomy=Static: Trained Policy2026.06 | 8.93 | 72.7 | 0.94 | |
| Always-Qwen3.5-PlusTaxonomy=Single-Model Baselines2026.06 | 2.68 | 80.1 | 0.19 |