ResearchBenchmarksCost/Accuracy Improvement Metrics on PopQA (test)Follow71.43Cost ReductionRISC61.144463.814766.48569.1553Jun 3, 2026Evaluation ResultsMethodMethodLinksCost ReductionAccuracy ImprovementHeadroom Over MaximumRISCBudget (@k LLM calls)=@10Budget (@k LLM calls)=@102026.0671.43——RISCBudget (@k LLM calls)=@15Budget (@k LLM calls)=@152026.0670——RISCBudget (@k LLM calls)=@5Budget (@k LLM calls)=@52026.0661.546.08—RISCBudget (@k LLM calls)=@50Budget (@k LLM calls)=@502026.06—4.754.31RISCBudget (@k LLM calls)=@75Budget (@k LLM calls)=@752026.06——5.34RISCBudget (@k LLM calls)=@99Budget (@k LLM calls)=@992026.06—6.776.42