Research
Baidu introduces an agent benchmark focused on finished work
Baidu's DuMateBench evaluates finished artifacts across 200 tasks, eight broad scenarios and 17 fine-grained categories.
Research.
Ideas, evidence and the questions shaping what comes next.
27 storiesPage 5 of 5
Research
China's daily AI token usage exceeds 500 trillion, MIIT confirms
An official September briefing confirms the June 2026 usage figure and distinguishes it from the much lower June 2025 baseline.
Research
CAICT report counts more than 70 operating embodied-AI training grounds
CAICT's August report describes more than 70 operating robot training grounds and over 40 planned or developing sites, alongside a new dataset-quality standard.
