ResearchBenchmarksMulti-hop reasoning on MuSiQue (val test)Follow2,849Token Usage (Optimization Phase)ETGPO2,013.567,652.7813,29218,931.22Feb 1, 2026Evaluation ResultsMethodMethodLinksToken Usage (Optimization Phase)ETGPOBackbone model=GPT-4.1...Backbone model=GPT-4.1-mini, Optimizer=GPT-4.12026.022,849GEPABackbone model=GPT-4.1...Backbone model=GPT-4.1-mini, Optimizer=GPT-4.12026.026,093MIPROv2Backbone model=GPT-4.1...Backbone model=GPT-4.1-mini, Optimizer=GPT-4.12026.0223,735