List your company

Multi-hop Question Answering on MuSiQue (EM, F1, Calls)

70.2EMReaLM-Retrieve
40.0447.8755.763.53Apr 29, 2026

Evaluation results

MethodLinks
ReaLM-RetrieveBackbone=DeepSeek-R1-671B2026.0470.277.81.9
Search-R1Backbone=DeepSeek-R1-671B2026.0465.873.42.6
IRCoTBackbone=DeepSeek-R1-671B2026.0464.271.83.6
ReaLM-RetrieveBackbone=DeepSeek-R1-Distill-Qwen-32B2026.0463.571.21.8
Search-R1Backbone=DeepSeek-R1-Distill-Qwen-32B2026.0459.1—2.4
Single RAGBackbone=DeepSeek-R1-671B2026.0458.766.21
IRCoTBackbone=DeepSeek-R1-Distill-Qwen-32B2026.0458.365.43.4
FLAREBackbone=DeepSeek-R1-Distill-Qwen-32B, Note=Uses token probabilities2026.0455.162.32.8
Self-RAGBackbone=DeepSeek-R1-Distill-Qwen-32B, Model Type=fine-tuned2026.0454.861.92.1
Single RAGBackbone=DeepSeek-R1-Distill-Qwen-32B2026.0452.659.41
No RetrievalBackbone=DeepSeek-R1-671B2026.0448.355.90
No RetrievalBackbone=DeepSeek-R1-Distill-Qwen-32B2026.0441.248.70