精选理由
开源模型GLM-5.2在AA-Briefcase智能体基准上只差闭源三个月了，做复杂任务时值得试试看。
Ethan Mollick根据Artificial Analysis的AA-Briefcase分数，绘制了AI模型在复杂多周咨询任务上的表现趋势。AA-Briefcase测试模型处理电子表格和策略规划等可交付成果。GLM-5.2等开源模型目前达到的水平，与闭源模型三个月前的分数一致。高端闭源模型仍保持明显领先，但差距在缩小。
原文 · @koltregaskes
Open weights are closing the time lag on tough age…
Open weights are closing the time lag on tough agentic benchmarks, 3 months now, GLM 5.2 is shrinking that gap.
Ethan Mollick graphed frontier performance on Artificial Analysis AA-Briefcase scores, which test models on multi-week consulting gigs with complex deliverables like spreadsheets and strategy work.
GLM-5.2 and similar recent open models hit scores closed models reached a few months earlier, though the absolute gap at the very top remains clear.