Chinese AI models are approximately 9-12 months behind US models by release timing, and the gap is potentially even larger when measured by time horizon, with some evidence suggesting Chinese models perform better on benchmarks than on held-out problems, possibly due to benchmark overfitting.