List your company

Multi-document Question Answering on MultidocQA

67.78HotpotQA AccuracyMemSearch-o1
46.449651.987357.52563.0627Apr 19, 2026

Evaluation results

MethodLinks
MemSearch-o1Backbone=Deepseek V3.12026.0467.7868.3252.01
MemSearch-o1Backbone=Qwen 2.5-72B-Instruct2026.0459.7165.9544.11
Search-o1 (RAgent)Backbone=Qwen 2.5-72B-Instruct2026.0457.4162.8644.78
AmberBackbone=Deepseek V3.12026.0455.5958.6341.29
Search-o1 (RAgent)Backbone=Deepseek V3.12026.0454.6456.1444.31
Direct RAGBackbone=Qwen 2.5-72B-Instruct2026.0454.447.2327.54
A-MemBackbone=Qwen 2.5-72B-Instruct2026.0454.2460.0839.95
Direct RAGBackbone=Deepseek V3.12026.0453.9843.7626.95
AmberBackbone=Qwen 2.5-72B-Instruct2026.0453.7861.1637.61
Search-o1 (Refined)Backbone=Qwen 2.5-72B-Instruct2026.0452.1135.7437.99
A-MemBackbone=Deepseek V3.12026.0449.5741.3825.67
Search-o1 (Refined)Backbone=Deepseek V3.12026.0448.4636.7533.32
MemoryBankBackbone=Deepseek V3.12026.0447.5146.6538.05
MemoryBankBackbone=Qwen 2.5-72B-Instruct2026.0447.2744.8928.26