Researchers have developed CORE, a new harness for deep research agents that addresses the trade-off between retrieving comprehensive search results and maintaining a concise agent context. CORE employs a coarse-to-fine search strategy, initially retrieving passages with a dense retriever and then using a late-interaction reranker to extract only the most relevant snippets. This approach, combined with context management to expand relevant sources and suppress duplicates, has demonstrated state-of-the-art accuracy and evidence recall across multiple agent models and benchmarks, outperforming existing harnesses like Tongyi DeepResearch and OpenResearcher. AI
IMPACT CORE's approach to context management and evidence retrieval could significantly improve the efficiency and accuracy of complex question-answering systems.
RANK_REASON The item describes a new research paper introducing a novel method for deep research agents. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
