Artificial Analysis launches AA-Briefcase benchmark; GPT-5.6 Sol tops Elo

AA-Briefcase evaluates models on multi-week, complex knowledge work projects built by industry experts. GPT-5.6 Sol (max) achieved the highest Presentation Elo, while Claude Fable 5 (max) was most efficient at 6.5 turns per task. Frontier models can take over 20 minutes per task.