{"version":1,"type":"story","url":"https://digestai.news/story/slice-mem-cuts-llm-memory-costs-by-99-5-in-new-framework","json":"https://digestai.news/story/slice-mem-cuts-llm-memory-costs-by-99-5-in-new-framework.json","markdown":"https://digestai.news/story/slice-mem-cuts-llm-memory-costs-by-99-5-in-new-framework.md","slug":"slice-mem-cuts-llm-memory-costs-by-99-5-in-new-framework","headline":"SLICE-Mem cuts LLM memory costs by 99.5% in new framework","summary":"Researchers introduced SLICE-Mem, a lightweight memory framework for LLM agents that reduces reliance on generative models for memory tasks. The system uses salience-guided admission and predicate-centered extraction to filter and organize user interactions without real-time LLM calls. Instead, it defers consolidation to an offline stage, cutting token consumption by **92.4%** on Memora Weekly and **99.5%** on LongMemEval-S compared to the baseline MemOS.\n\nEvaluated on subsets of Memora and LongMemEval-S, SLICE-Mem achieved **46.60** FAMA on Memora Weekly and **56.8%** accuracy on LongMemEval-S using **8B** and **30B** Qwen backbones. While slightly trailing MemOS’s **46.68** and **59.3%**, the framework demonstrates efficiency by minimizing LLM inference overhead while maintaining coherence in long-term agent interactions.","keyPoints":["SLICE-Mem uses salience-guided admission and predicate extraction to filter user turns without real-time LLM calls","Reduces token consumption by **92.4%** on Memora Weekly and **99.5%** on LongMemEval-S vs. MemOS baseline","Achieved **46.60** FAMA on Memora Weekly and **56.8%** accuracy on LongMemEval-S with Qwen backbones"],"whyItMatters":"SLICE-Mem addresses a key bottleneck in LLM agents: high inference costs for memory management. By cutting token usage by nearly 100% in some cases, it could enable cheaper, scalable long-term memory for agents without sacrificing accuracy.","category":{"slug":"research","name":"Research","url":"https://digestai.news/category/research"},"entities":{"companies":[],"models":["SLICE-Mem","MemOS","Qwen"],"people":[]},"firstPublishedAt":"2026-10-27T04:00:00Z","updatedAt":"2026-10-27T04:00:00Z","sourceCount":1,"hasPrimarySource":true,"sources":[{"outlet":"arXiv cs.CL","title":"SLICE-Mem: Salience-Gated Lightweight Information Curation for Efficient LLM Agent Memory","url":"https://arxiv.org/abs/2610.33015","publishedAt":"2026-10-27T04:00:00Z","type":"primary","primary":true,"lead":true}],"sourceNotes":null,"discussions":[],"thread":{"title":"The Rise of External Memory in AI Agents","url":"https://digestai.news/thread/tencent-releases-hy-memory-30b-a3b-for-external-memory-frameworks","storyCount":3},"cite":{"text":"Digest AI, \"SLICE-Mem cuts LLM memory costs by 99.5% in new framework\", 27 October 2026, https://digestai.news/story/slice-mem-cuts-llm-memory-costs-by-99-5-in-new-framework","publisher":"Digest AI","title":"SLICE-Mem cuts LLM memory costs by 99.5% in new framework","datePublished":"2026-10-27T04:00:00Z","url":"https://digestai.news/story/slice-mem-cuts-llm-memory-costs-by-99-5-in-new-framework"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}