Predictive LLM Routing on HELM Lite benchmark leaderboard · SOTA2 Research

Predictive LLM Routing on HELM Lite

64.3OpenAI PerformanceOracle

Evaluation Results

Method	Links
Oracle2025.05		64.3	63.75	66.74
Linear2025.05		55.61	53.54	52.64
MLP2025.05		55.6	53.21	51.46
Linear2025.05		55.31	53.4	51.93
kNN2025.05		54.73	53.4	52.18
MLP2025.05		54.43	50.82	51.51
kNN2025.05		53.93	52.66	50.92
Graph2025.05		53.34	45.19	50.45
D-Attn2025.05		53.31	50.17	51.38
Attn2025.05		53.07	52.35	51.89
Graph2025.05		52.7	51.89	51.41
Attn2025.05		52.23	52.12	50.61
D-Attn2025.05		51.49	49.43	51.82
Graph2025.05		50.47	44.62	50.72
kNN2025.05		49.75	44.85	49.48
MLP2025.05		49.3	45.37	48.95
D-Attn2025.05		49.29	42.06	48.33
Attn2025.05		49.13	44.31	49.17
D-Attn2025.05		48.95	40.76	48.6
Random2025.05		48.94	41.89	45.15
kNN2025.05		48.93	41.86	48.63
Graph2025.05		48.84	43.11	48.55
MLP2025.05		48.6	43.55	48.72
Linear2025.05		48.59	42.57	48.73
Linear2025.05		48.53	42.13	48.69
Attn2025.05		48.45	42.08	48.04