Multi-objective preference selection on OpenAI models Setup A1: K=5, M*=2 benchmark leaderboard · SOTA2 Research

Multi-objective preference selection on OpenAI models Setup A1: K=5, M*=2

0.055Cost ($)ϵ-greedy

Evaluation Results

Method	Links
ϵ-greedy2026.05		0.055	2.5	80
CUPID2026.05		0.056	2.2	100
CUPID without language2026.05		0.06	4.2	100
CUPID without cost penalty2026.05		0.081	4.2	100
Cost-constrained LMA2026.05		0.091	5.75	80
Latent Dueling Bandits2026.05		0.108	—	0
LMA2026.05		0.115	5	60
RUCB2026.05		0.161	—	0