ResearchBenchmarksLLM Evaluation Efficiency on 100k sessionsFollow4Task CostBoRP-22.92158.79340.5522.21Jan 26, 2026Evaluation ResultsMethodMethodLinksTask CostThroughput (sessions/hour)Relative Cost (%)BoRPModel size (parameters...Model size (parameters)=8B, Hardware (NVIDIA A100)=one A1002026.01457,8000.6GenModel size (parameters...Model size (parameters)=14B, Hardware (NVIDIA A100)=one A1002026.01287,3004.1Qwen3-MaxAccess method=APIAccess method=API2026.01124—18.3GPT-4oAccess method=APIAccess method=API2026.01677——