Moonshot AI Cuts Kimi K3.1 Pro Subagent Calls After Quota Complaints
Moonshot AI said it would compensate users affected by rapid quota consumption in Kimi K3.1 Pro and reduce unnecessary subagent calls and output length. The company acknowledged that the issue should have been fixed earlier, following reports that small tasks could consume five hours of quota. Anthropic separately advised folding subagents into the main agent loop when one skilled agent can handle a task.
The incident shows that excessive subagent orchestration can make agent products costly and unpredictable for subscription users, while Anthropic’s guidance presents single-agent execution as a possible remedy.
K3.1 Pro发布之后，由于多位用户反馈额度烧得太快，月之暗面紧急发布公告，称“问题出在我们，我们也本应更早修复”，在补偿额度的同时，月之暗面也表示将“减少不必要的子Agent调用和输出长度”。
就在模型上线前两天，Anthropic发了一篇文章，标题为《构建Agent：何时该把子Agent折叠回主循环》（Building agents: When to fold subagents into your main agent loop），它所讲的，刚好就是关于子Agent这件事。
Anthropic的结论是，通过对比多家企业部署的Claude他们发现，一个带技能的Agent，质量反而胜过了“一群子Agent分工”，而且单任务成本和延迟往往更低。最关键的是，这个Agent所消耗的token，要远远低于一群子Agent。
子Agent这个概念，其实是模型能力不足时期，厂商为了让模型完成更多任务所打的一个“补丁”。
从任务完成率来说，多个子Agent确实要远远强于当时的单Agent。作为代价，token烧得更厉害了，模型也就变得更贵了。
#Moonshot AI#Kimi K3.1 Pro#Anthropic#Claude#single-agent-loop-kit