ResearchBenchmarksQuestion Answering on CWQ, HotpotQA, Quasar-T, and StrategyQA Average (test)For AI assistantsMarkdownContent indexQuestion Answering on CWQ, HotpotQA, Quasar-T, and StrategyQA Average (test)Follow1.36Inference Time (s)PRAG1.0438283.1779895.312157.446311Jun 12, 2026Evaluation resultsMethodMethodLinksInference Time (s)GPU Memory Usage (GB)PRAG2026.061.364.83SFT-LORA2026.061.674.82BASIC-RAG2026.061.892.54DMoE2026.062.66567.2352FLARE2026.069.264313.9718