Multi-hop Question Answering on MARCH benchmark leaderboard · SOTA2 Research

Multi-hop Question Answering on MARCH

73Exact Match (EM)Human

Evaluation Results

Method
Human2025.09	73	62	67.5	—
CLARION2025.09	38.73	28.38	33.56	3.474
CLARION2025.09	31.47	27.03	29.25	3.042
CLARION2025.09	29.12	26.3	27.71	2.752
DIVA2025.09	28.82	22.73	25.78	3.015
CoT w/ RAG2025.09	25.63	26.61	26.12	2.947
NaiveRAG2025.09	25.1	26.2	25.65	2.752
CLARION w/o clarification2025.09	25.1	25.56	25.33	2.922
CLARION w/o clarification & detection2025.09	24.51	24.31	24.41	2.906
CLARION w/o clarification2025.09	24.12	22.54	23.33	2.609
CLARION w/o clarification & detection2025.09	23.82	22.04	22.93	2.573
CLARION w/o clarification2025.09	23.63	22.99	23.31	2.927
ReAct2025.09	23.17	24.78	23.97	2.723
CoT w/ RAG2025.09	23.15	27.31	25.23	2.373
CLARION w/o clarification & detection2025.09	22.94	24.02	23.48	2.782
NaiveRAG2025.09	22.16	28.63	25.4	2.297
CoT2025.09	21.51	22.32	21.91	2.897
CoT w/ RAG2025.09	21.33	23.18	22.25	2.632
ReAct2025.09	21.32	22.37	21.84	2.428
No Retrieval2025.09	20.98	21.19	21.09	3.083
ReAct2025.09	20.98	21	20.99	2.832
NaiveRAG2025.09	20.2	25.03	22.62	2.084
CoT2025.09	19.8	22.12	20.96	2.512
DIVA2025.09	18.82	20.29	19.56	2.303
DIVA2025.09	18.82	20.66	19.74	2.636
No Retrieval2025.09	17.75	18.72	18.24	2.683
CoT2025.09	16.32	17.52	16.92	2.258
No Retrieval2025.09	15.59	20.1	17.85	2.307