List your company

LLM Routing on Cold-start

0.0226Cost (Cost-first)Smallest LLM
0.0111120.0886560.16620.243744Apr 10, 2026

Evaluation results

MethodLinks
Smallest LLMEvaluation Protocol=LLM-as-a-judge2026.040.02260.33420.04880.02260.33420.15580.02260.33420.26290.4675
Adaptive LLMEvaluation Protocol=LLM-as-a-judge2026.040.02260.33420.04880.06370.50240.21940.30820.51580.3510.6191
Smallest LLM2026.040.0230.20040.02170.0230.20040.08870.0230.20040.15570.2661
Adaptive LLM2026.040.0230.20040.02170.06410.4260.18090.30980.43830.28870.4913
MetricRouterCross-domain=true2026.040.03250.19160.01230.04460.20270.07030.10120.21180.14920.2319
TRouterSynthesis framework=gemini-2.5-flash2026.040.03270.30680.03520.05260.41440.18090.07010.42010.32210.5382
RouterDCCross-domain=true2026.040.03910.25480.01970.05990.3580.1490.0650.38990.29890.4676
Prompt LLMEvaluation Protocol=LLM-as-a-judge2026.040.03940.40530.04950.10880.48530.18830.30820.51580.3510.5888
Prompt LLM2026.040.04270.21310.00850.11280.39960.14340.30980.43830.28870.4406
TRouterEvaluation Protocol=LLM-as-a-judge2026.040.06170.49920.05050.06370.5030.21960.0740.49970.3850.6551
MetricRouterEvaluation Protocol=LLM-as-a-judge, Cross-domain Training=true2026.040.0620.49670.04980.1240.53970.20780.18330.54930.1830.4406
TRouterSynthesis framework=gpt-4.12026.040.06210.42570.03550.06410.42620.18110.17210.43150.31080.5274
RouterDCEvaluation Protocol=LLM-as-a-judge, Cross-domain Training=true2026.040.07130.55730.05440.09140.52310.21590.11480.52970.20740.4777
Largest LLMEvaluation Protocol=LLM-as-a-judge2026.040.30820.51580.14340.30820.51580.10380.30820.51580.3510.3113
Largest LLM2026.040.30980.4383-0.16010.30980.43830.06430.30980.43830.28870.1928