ResearchBenchmarksLong-context multi-hop evaluation on MuSiQue-Long (3.2k tok)Long-context multi-hop evaluation on MuSiQue-Long (3.2k tok)Follow37.7EMTUR-DPO28.54830.92433.335.676Apr 30, 2026Evaluation resultsMethodMethodLinksEMPath CoverageCyclesTUR-DPO2026.0437.761.89.3DPO2026.0433.654.214.1SFT2026.0428.948.718.5