LLM Routing Resource Efficiency on LLM Routing Systems
0GPU Memory Usage (GB)RouteLLM (MF)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| RouteLLM (MF)Model=Mat. factorization2026.03 | 0 | — | |
| RouteLLM (BERT)Model=0.3B classifier2026.03 | 0.6 | — | |
| vLLM Semantic RouterModel=3× mmBERT-32K2026.03 | 0.8 | — | |
| NVIDIA BlueprintModel=BERT/CLIP + Triton2026.03 | 1 | — | |
| RouteLLM (Causal)Model=8B LLM2026.03 | 16 | — | |
| R2-RouterModel=Reasoning LLM2026.03 | 16 | — |