---
id: 20260508-T0-09
title: "LCM架构在长编码任务上超越Claude Code"
title_en: "LCM Architecture Outperforms Claude Code in Long Context Tasks"
url: https://ai.daily.yangsir.net/daily/20260508-T0-09
issue_date: 2026-05-08
publish_date: 2026-05-07T04:00:00.000Z
category: research
source_name: "arXiv cs.AI"
source_url: https://arxiv.org/abs/2605.04050
---

# LCM架构在长编码任务上超越Claude Code

研究团队提出无损上下文管理架构LCM，经测试显示其性能优于Claude Code。基于该架构的编码助手Volt在Opus 4.6评测中获得更高分数。LCM采用确定性设计，解决了大模型长文本处理的效率问题。开发者可通过该技术提升AI在复杂编程任务中的表现。

## English Version

**LCM Architecture Outperforms Claude Code in Long Context Tasks**

Researchers introduced Lossless Context Management (LCM), a deterministic architecture for LLM memory that outperforms Claude Code on long-context tasks. The LCM-augmented coding agent Volt achieves higher scores than Claude Code in Opus 4.6 benchmarks. LCM solves efficiency problems in long-context processing for AI applications.

---

**来源**：[arXiv cs.AI](https://arxiv.org/abs/2605.04050)

**详情页**：https://ai.daily.yangsir.net/daily/20260508-T0-09

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*