SemiAnalysis reviewed 857 models released by nine Chinese developers from 2021 through September 15, 2026. It found specific model-level safety results published for 31, or 3.6%, and results published before or at release for nine, or 1.1%. For 16 models with confirmed post-release disclosures, the median delay was 42 days. General statements that safety training had occurred were not counted as results, and failure to find public material does not mean no internal testing occurred. Vendors also count model variants differently, so the transparency of each model’s evaluation matters more than a simple ranking.
Back to Tech
Review of 857 Chinese AI models finds published safety results for 31
Missing public evidence is not proof of missing internal tests; adoption decisions should check model-specific evaluations and disclosure dates.