Understand this story
On October 24, 2026, a study proposed RAW-RAG, which performs only dense retrieval plus reranking over raw conversation history, and found it stronger than complex agentic memory on long-conversation memory benchmarks. The study compared on four benchmarks—LoCoMo, LongMemEval, MEME, and MemoryAgentBench—and RAW-RAG ranked first in 6 of 7 benchmark-and-model combinations, scoring 0.13 to 5.53 points higher than the strongest agentic memory system. The authors argue the bottleneck lies mainly in whether evidence is complete, rather than in extraction, mutation, and graph structure; ablations show that removing reranking drops the score by 3.92 points and returning only 1 item drops it by 11.73 points, while adding atomic fact extraction, summary augmentation, or graph expansion all failed to exceed the default RAW-RAG. The authors self-report limitations: it covers only offline public conversation benchmarks, does not involve interaction, multimodality, or real-world operations, and the privacy and deletion constraints of raw-text storage are a real-world cost.
AI-generated from public reports
Latest development · Report
Research shows that naive RAW-RAG outperforms complex Agentic Memory on long-conversation memory benchmarksDevelopments
1 confirmed developments
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Oct 24
Shao Meng
Research shows that naive RAW-RAG outperforms complex Agentic Memory on long-conversation memory benchmarks
A study compared on four long-conversation memory benchmarks—LoCoMo, LongMemEval, MEME, and MemoryAgentBench—and found that RAW-RAG, which only performs dense retrieval plus reranking on the raw conversation history, ranked first in 6 of 7 benchmark-and-model combinations, scoring 0.13 to 5.53 points higher than the strongest agentic memory system. The authors believe the bottleneck lies mainly in whether the evidence is complete, rather than in extraction, mutation, and graph structure; ablations show that removing reranking drops performance by 3.92 points, and returning only 1 item drops it by 11.73 points, while adding atomic fact extraction, summary augmentation, or graph expansion all fail to surpass the default RAW-RAG. The authors state their limitations as covering only offline public conversation benchmarks, not involving interaction, multimodality, or real-world operations, and the privacy and deletion constraints of raw-text storage are a real-world cost.