ResearchBenchmarksQuestion Answering on HotpotQA (test) (Cost Reduction and Improvement Metrics)Follow64.29Cost ReductionRISC36.428443.661750.89558.1283Jun 3, 2026Evaluation ResultsMethodMethodLinksCost ReductionAccuracy ImprovementHeadroom Over MaximumRISCBudget (@k LLM calls)=@15Budget (@k LLM calls)=@152026.0664.29——RISCBudget (@k LLM calls)=@10Budget (@k LLM calls)=@102026.0650——RISCBudget (@k LLM calls)=@5Budget (@k LLM calls)=@52026.0637.55.95—RISCBudget (@k LLM calls)=@50Budget (@k LLM calls)=@502026.06—6.494.79RISCBudget (@k LLM calls)=@75Budget (@k LLM calls)=@752026.06——6.63RISCBudget (@k LLM calls)=@99Budget (@k LLM calls)=@992026.06—8.037.98