Open-weight AI caught the frontier on capability. On safety, it didn’t

Open-weight AI models have nearly caught the frontier on capability. And once the weights are public, no lab can enforce a guardrail. A new evaluation of China’s leading open model makes the gap concrete.