Compact-Bench：衡量上下文压缩对长程编码智能体的代价
原标题：Compact-Bench: Measuring What Context Compaction Costs Long-Horizon Coding Agents
AI 摘要
该论文提出 Compact-Bench 基准，用于衡量上下文压缩（context compaction）对长程编码智能体续作能力的影响。它从同一仓库状态出发，对比完整历史与仅摘要续作，覆盖 SWE-bench 修复、智能体会话和生产链。结果显示，稳定损伤范围为完整历史成功案例的 10%–20%，Claude Opus 保留最多修复，DeepSeek-Flash 损伤最大；以 Kimi-K3 为摘要器时，Kimi Code 提示保留最多任务续作，Claude Code 保留最多重复细节。论文指出，仅保留相关信息不足以保证任务成功续作。
正文节选
Compact-Bench: Measuring What Context Compaction Costs Long-Horizon Coding Agents Abstract Context compaction lets long-horizon coding agents continue beyond their context windows, but can erase the state needed to finish a task. Prior evaluations measure efficiency, retained information, or end-to-end success, leaving the effect of a specific history replacement difficult to isolate. Compact-Bench pairs full-history and summary-only continuations from a shared repository state, and evaluates re