Opus 5.5 matches human baselines in CMU AI infrastructure benchmark
新智元 reports that a Carnegie Mellon University benchmark evaluated 11 agents on six end-to-end AI infrastructure tasks spanning pretraining, intermediate and post-training, inference, agent runtimes, and cluster orchestration. Opus 5.5 paired with Claude Code reached or exceeded the human baseline on all six tasks, with the reported systems matching or outperforming production tools such as SGLang and NVIDIA Dynamo on the same GPUs. The source also says Opus 5.5 scored 1.25 on the Singularity-Index, while GPT-6.1 Sol scored 0.969 and the other agents averaged 0.58.
The reported results position coding agents as competitors in building and optimizing production-grade AI infrastructure, a capability traditionally associated with specialist human engineering teams.

你让AI帮忙修一个登录报错。照着它的建议改完代码,你重新跑了一遍登录测试。终端又弹出一行红字,报的还是同一个错。你把新日志贴进对话框,再按下回车,等AI告诉你下一步查哪儿。修一个bug,往往要这样来回折腾几轮。每次提交问题后,你要等多久,才能看到AI回的第一个字?最近,有人拿真实编程会话做了个测试,Claud......
本文来自微信公众号: 新智元 ,编辑:元宇,作者:ASI启示录,原文标题:《Claude自写推理引擎!等AI回第一个字,12秒砍到半秒》
你让AI帮忙修一个登录报错。
照着它的建议改完代码,你重新跑了一遍登录测试。终端又弹出一行红字,报的还是同一个错。
#Opus 5.5#Claude Code#Singularity-Bench#Carnegie Mellon University#SGLang#NVIDIA Dynamo#GPT-6.1 Sol