Artificial Analysis 发布新智能体知识工作基准测试 AA-Briefcase，用于评估模型在复杂项目中的真实知识工作能力。GPT-5.6 Sol (max) 在该基准中取得最高 Presentation Elo，其生成的 PowerPoint 和 Excel 等文件输出在专业呈现上超越所有其他模型。相比前代 GPT-5.5 (xhigh)，GPT-5.6 Sol (max) 的 Presentation Elo 提升约 500 分，在模型间视觉对比中预计胜率约 95%。AA-Briefcase 评分结合二元 Rubric 检查与 Analytical Quality 及 Presentation 两两比较。
GPT-5.6 Sol (max) has the highest Presentation Elo of any model in AA-Briefcase
AA-Briefcase is our new agentic knowledge work benchmark for testing models on realistic knowledge work tasks in complex projects built by industry experts. Grading in AA-Briefcase combines binary Rubric Checks with Analytical Quality and Presentation pairwise comparison. GPT-5.6 Sol (max) has the highest recorded Presentation Elo - its outputs across various file types, including PowerPoint and Excel, are the most professionally presented of any model.
This is a significant jump from the model’s predecessor. GPT-5.6 Sol (max) gains ~500 Presentation Elo points on GPT-5.5 (xhigh) - resulting in a projected win rate of ~95% in head-to-head visual comparisons between the models.
来源：Artificial Analysis · x.com