Kimi 发布 PerceptionBench 视觉感知基准
We are releasing PerceptionBench, a benchmark that isolates visual perception and evaluates it as a set of atomic capabilities - discovered from how today's mod
我们发布 PerceptionBench，一个隔离视觉感知、并将其作为一组原子能力进行评估的基准——这些能力是从当今模型的失败方式中发现的，而非预先定义。 从前沿模型在 42 个基准上的失败中，我们推导出 10 项原子感知能力，并构建了 3,000 道经过验证的问题，每道题只隔离一项能力，仅凭观察即可作答，无需推理或外部知识。
正文
当前显示原文，中文全文译文尚未就绪。
We are releasing PerceptionBench, a benchmark that isolates visual perception and evaluates it as a set of atomic capabilities - discovered from how today's models fail, rather than defined in advance.
From frontier-model failures across 42 benchmarks, we derive 10 atomic perceptual capabilities and construct 3,000 verified questions, each isolating a single capability and answerable by looking, with no reasoning or external knowledge required.