LangChain 在 Open SWE 中验证路由器置于 harness 内可省 64% 成本
router lives in the harness, not a gateway, bc the harness knows the task blog: https://t.co/k9b306...
LangChain 给 Open SWE 编程智能体装了个路由器，973 个线程测下来每个省 2/3 的钱，成功率还没掉，做法细节都公开了。
LangChain 在开源编程智能体 Open SWE 中测试了模型路由方案，将路由器放在 harness 层而非网关层，由 harness 依据任务信息做路由。在 973 个线程的对比测试中，路由组的线程中位成本为 0.94 美元，对照组全部使用 GPT-6 Astra 为 2.61 美元，成本降低 64%。两组最终合并 PR 的比例分别为 29.2% 和 27.3%，p 值 0.49，说明成功率差异不显著。路由器将 56% 的线程分给均衡模型、34% 给快速模型、10% 给性能模型；全程只用快速模型的方案因输出质量低在一天内被放弃。LangChain 自行指出该测试的对照是最贵模型，路由器目前仍是概念验证。
router lives in the harness, not a gateway, bc the harness knows the task blog: https://t.co/k9b306...
router lives in the harness, not a gateway, bc the harness knows the task blog: langchain.com/blog/how-to-bu… x.com/neil_xbt/statu… NeilXbt @neil_xbt LangChain gave its coding agent three models and a router. Open SWE, their open-source coding agent, now sends each thread to the cheapest model likely to finish the job. they tested it on 973 threads, half routed and half always on GPT-6 Astra. median cost per thread: $0.94 against $2.61. that's 64% lower. threads ending in a merged PR: 29.2% routed, 27.3% on Astra. LangChain's p-value for that gap is 0.49. the router sent 56% of threads to the balanced model, 34% to the fast one and 10% to the performance one. they also tried always using the fast model. that test ended within a day, because engineers reported low output quality. LangChain's own caveats: the control was the most expensive model, and the router is a proof of concept. Which three models would you put in the tiers? 🔗 View Quoted Tweet 💬 10 🔄 3 ❤️ 20 👀 1779 📊 11 ⚡
