arXiv 2610.28577 · cs.CL · 2026-10-20
Cache Engineering for Agents
Tianyu Fu, Ruoxi Wang, Zihan Min, Zhongqi Ren, Zhihao Li, Tengxuan Liu, Shengxu Mu, Hanling Zhang and 1 more
Tsinghua-NICS-EFC
In latest · #40 Added on 2026-10-23
One-line takeaway
Tsinghua researchers propose cache engineering: agents directly edit their KV cache (drop entries, reposition keys) instead of trimming or summarising history, avoiding re-prefill; with the co-designed AKV engine, a rolling-drop policy cuts average cost by 27.7% and adds 1.7 benchmark points vs summarisation.
清华团队提出“缓存工程”：让智能体直接操作自己的 KV Cache（drop 丢弃、repos 调整位置编码），而不是裁剪或总结历史，避免重新预填充；配套 AKV 推理引擎，相比阈值触发的摘要方案平均成本降 27.7%、基准分高 1.7 分。
Read on arXiv PDF Hugging Face discussion Code Project page
Topics & disciplines
More on Agents
- Raven: The Harness of Harnesses for Composable Agentic Intelligence ▲748
- False Frontiers: Diagnosing and Mitigating Co-Cheating in Self-Evolving Search Agents ▲642
- Omni-IO Skills: Harnessing Your Agent Omni-Native ▲485
- BizAgentBench: Unveiling Hidden Challenges in Deploying Long-Horizon Agents for Real-World Business ▲388
- Player-1: Learning to Play from Synthetic Experience ▲347
More on LLMs
- Jeva: System One Models from First Principles 2026-10-13 · ▲371
- Post-Training Leaves Behavioral Shadows on Unrelated Decisions 2026-09-24 · ▲273
- WeVisOne-VL: Unified Visual Perception from Objects to Pixels 2026-10-22 · ▲247
- Thinking Inertia: LLMs Keep Thinking When Told Not To 2026-10-08 · ▲241
- OmniEdu: Open Foundation Models for Learning and Teaching 2026-09-19 · ▲240
More from Tsinghua-NICS-EFC
About this data
Metadata from the arXiv API (verified); upvotes from Hugging Face Daily Papers; citations from Semantic Scholar. Last refreshed 2026-10-26.
The takeaway is written by AIMap from the abstract; the abstract itself is not reproduced — read the original.