MMLU & MMLU-Pro
基准测试
能力
The default knowledge exam of 2020–24 — now saturated.
您正在查看一条基准测试记录：这是业界用于衡量芯片或模型的测试——以及它实际告诉你什么。
MMLU's 57-subject multiple-choice exam was the standard capability score from GPT-3 (43.9%) to the high-80s frontier plateau; MMLU-Pro hardened it as scores saturated. Its arc is the pattern every benchmark follows: introduced, climbed, saturated, replaced.
它如何与整个技术栈相连
本记录的关联链接一览——点击分组可直接跳转。
相关
来源
记录审阅截至 2026-01-15 — 查看 方法论.