Moonshot shipped a weaker model on purpose. Kimi K3.1 Pro's exploit test fell … · @freestyle_error · TwKit

@freestyle_error

Moonshot shipped a weaker model on purpose. Kimi K3.1 Pro's exploit test fell from 90.5 to 81.0 once cyber data was removed. Terminal-Bench: 65.7, above GPT-6 Astra. Open weights cannot be recalled.

曝光 30 · 评论 1 · 点赞 3 · 书签 0