AI 导读
过去几周，我们做了非正式实验，测试 GPT-5.6 Sol 的 computer use，让它玩电子游戏《Slay the Spire》。 我们发现 computer use 并不是 Sol 的主要限制。Sol 主要卡在战术和策略上。它玩得比人类新手好，但低于专家水平。
正文
Over the last few weeks, we ran informal experiments testing GPT-5.6 Sol's computer use by having it play the video game Slay the Spire.
We learned that computer use is not Sol's main limitation. Rather, Sol mostly struggles with tactics and strategy. It plays better than new human players but below expert level.
来源：@EpochAIResearch · x.com