Opus 5狂烧6.9亿token做游戏，GPT-5.6用5美元复刻了
看GPT-5.6怎么用五美元复刻Opus 5几亿token做出来的游戏，差距是真大。
Opus 5为制作一款游戏烧掉了6.9亿token。GPT-5.6仅用5美元成本就复刻了这款游戏。两者在token消耗和花费上差距悬殊，显示出GPT-5.6在游戏生成任务上的成本效率优势。
Opus 5狂烧6.9亿token做游戏，GPT-5.6用5美元复刻了
据说GPT-5.6很适合拿来做游戏
博主实测说 DeepSeek 4.1-1017 打平 Opus 5，还压过 GLM5.3 和 GPT-5.6 Sol，国产模型又卷起来了。
Evolvent AI 把递归自我改进研究环境开源了，连 Opus 5 绕过预算偷调 14 个模型的日志都放出来，能看 Agent 怎么钻空子。
This reframes RSI as a systems engineering challenge, highlighting how environment design impacts research agent effectiveness. The platform's Everything as a Service approach could accelerate AI development cycles for companies building full-stack AI systems. Developers can access the code and experiment configurations to test their own improvement methods.
这篇论文测试了让智能体自己动手省钱：让 Opus 5 等模型把任务封装成小程序或小模型，成本能降 657 倍但多数尝试会翻车，结果挺有意思。
The finding inverts a common architecture assumption: giving every user their own agent can be worse than centralising one agent that holds all constraints. Builders designing multi-agent systems should test shared-resource scenarios before committing to per-user agents, or make reading peer agents a precondition for any action. The paper benchmarks 5 frontier models including Opus 5 and Claude, giving a concrete baseline for comparing your own stack.
