Why are Chinese labs so focused on open models?

Chinese labs appear more inclined to release open-weight AI models, contrasting with US labs that tend to keep their models closed. This suggests Chinese researchers might not view model weights as the primary competitive advantage in AI, speculating that specialized training data or evaluations could be the true differentiator. The observation that some expert training data originates from US companies like Mercor/SurgeAI adds an interesting layer to this dynamic, or it could simply reflect a cultural difference.

This report uniquely speculates on the underlying reasons for Chinese labs' preference for open-weight models, unlike the more common focus on their existence.