ResearchBenchmarksCode Generation on HumanEval-X PythonFollow0.02Cost (cents)Otrue0.0080.0890.170.251Jun 25, 2026Evaluation ResultsMethodMethodLinksCost (cents)Cost Saving (%)OtrueSelection Strategy=Opt...Selection Strategy=Optimal router, Alpha=alpha*2026.060.0294.8SOTA LLMSelection Strategy=Glo...Selection Strategy=Globally best single LLM2026.060.32—