Agent37 用 CPU 共享设计跑 100 个编码智能体,每小时仅 $0.68
编码 agent 大部分时间在等模型,CPU 只跑满约 5%,Agent37 干脆让 100 个 agent 共享物理核,价格压到每小时 $0.68。批量跑任务的人可以算下这笔账,单任务追求稳定响应的则要掂量共享带来的干扰。
Agent37 推出一种针对编码智能体的运行方案,让 100 个 agent 共用一套物理资源,整体成本降到每小时 $0.68。核心依据是一个观察:编码 agent 的 CPU 大部分时间处于空闲,真正跑满的时间大约只有 5%,瓶颈在等待模型响应上。
基于这个特点,Agent37 放弃了给每个 agent 分配独占 CPU 的传统做法,改用多 agent 共享同一批物理核心,把原本闲置的算力利用起来。这是一种用 CPU 隔离性换价格的设计取舍:单个 agent 的响应可能受邻居任务影响,但整体成本大幅下降。
对需要批量跑编码任务(如 CI 修复、批量重构、多分支测试)的团队,这类共享型运行时可能明显降低 token 之外的算力账单;对延迟敏感的单任务场景,独占资源方案仍有优势。
And a coding agent may use its CPU only about 5% of the time, since it mostly waits on the model, so Agent37 lets many agents share the same physical cores.
And that sharing lets it run 100 agents for $0.68 an hour. so this design trades dedicated CPU for price
This report is categorized under AITOP's "Products" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include Agent, coding assistants, cost optimization, Agent37.
Source attribution: curated from rohanpaul_ai (@rohanpaul_ai).