NVIDIA：智能体无需长上下文也能完成长程任务
"You Don't Need Long Context"
摘要
NVIDIA 发现 LLM 智能体无需保留完整对话历史即可完成复杂长程任务：通过频繁的上下文压缩总结进展、按需检索信息，甚至在每次动作后重置上下文。在六个长程基准上，该方法保持准确率的同时将估计推理成本最多降低 8.8 倍。
摘要由模型根据原文撰写，以原文为准。
正文 · 原文
"You Don't Need Long Context"
NVIDIA finds that LLM agents can complete complex long-horizon tasks without maintaining their entire conversation history, which reduces inference costs.
So instead of accumulating tokens, they use frequent context compaction to summarize progress and retrieve information when needed, even resetting the context after every agent action.
Across six long-horizon benchmarks, this preserves accuracy while achieving up to 8.8x lower estimated inference costs.
来源：alphaXiv · x.com
引用原文
askalphaxiv. "You Don't Need Long Context"[EB/OL]. (2026-11-06)[访问日期]. https://x.com/askalphaxiv/status/2118576609603252583.