Open Models Close the Gap, But the West Relies on Eastern Innovation

The debate over open and closed models has shifted from ideology to pure economics. According to analyses, the gap between the closed frontier and open-source has shrunk from years to just a few months. The main driver of this acceleration is Chinese labs, which effectively use techniques like distillation to train their own high-performing open models.

Corporations swap patriotism for lower invoices

This shift has concrete business implications. American and European companies are increasingly replacing expensive models like Claude or GPT with cheaper Chinese open-weights alternatives like DeepSeek or Qwen. This is not a political decision, but a move to reduce infrastructure costs.

Distilling synthetic data hits the boundaries of bans

While Western discourse often focuses on banning distillation and protecting intellectual property, real research shows that synthetic data from large models is a necessity for any further progress by smaller labs.

Data does not respect geopolitical boundaries

Distillation is not just copying; it is an optimization that all players are now using for faster and cheaper training.

Market loss won't be reversed by regulatory statements

Any attempts to impose a blanket ban on powerful open models are likely to hit the wall of these tools being already too deeply integrated into the ecosystem. The real signal will not be new political statements, but whether American labs can find an economic model to keep up with Asian competition.

Lilith's verdict

The West is gearing up for a war over intellectual property, while Asia has long since cheapened the model and sold it to corporations as an infrastructure commodity. You can ban models in Congress, but you will have a hard time banning them in the budgets of CTOs.