谷歌联合 Artificial Analysis 推出 MKA 基准测试：Gemini 4 Carbon 多模态知识工作测试获第二，任务成本仅为第一名六分之一Chinese only

Google and Artificial Analysis launch MKA benchmark for multimodal knowledge work

MKA uses 200 tasks across industries to assess how AI agents handle visual files and complete knowledge work. The official test tasks are private, while 10 other tasks are public. In results published by Google, Gemini 4 Carbon ranked second for quality at about one-sixth the average task cost of the top-ranked model. Developers can now evaluate models through Artificial Analysis.