PowerContext v1.5.0 targets token overhead in long-running agents
OceanBase released PowerContext v1.5.0 with reusable prepared context, reversible shared-JSON encoding, evaluation tools, and an alpha Rust Gateway for long-running agent workflows. In the publisher’s paired SWE-bench Pro comparison, the integration reportedly reduced cumulative input tokens by 25.4% without lowering task-completion performance. The roughly 75% reduction cited in the headline applies to a single-task input peak, not the overall benchmark total.
The release targets a major cost driver in long-running software-engineering agents by reusing context and avoiding repeated inputs, although its benchmark savings remain publisher-reported.
Agent 处理一个跨模块缺陷，通常需要多轮读取代码、修改实现和运行测试。进入下一轮时，项目约束、历史事实和已经确认的线索仍然需要保留，相关的检索、上下文组装和模型输入也会持续产生开销。
PowerContext v1.5.0 将 token 效率工程作为本次更新的重点之一。 具体做了三件事：复用已经准备好的上下文，减少记忆与经验处理中的重复输入，并提供评测工具，检查这些优化是否降低了完成任务的实际消耗。
在本次 SWE-bench Pro 配对对照中，PowerContext 集成方案在任务完成效果未下降的情况下，累计输入 token 减少约 25.4%。
#OceanBase#PowerContext v1.5.0#SWE-bench Pro#Scale AI#Rust Gateway