The Chinese Models are Winning

We know that whilst OpenAI and Anthropic and battling it out for revenue growth to justify their insatiable demand for capital, the Chinese are doubling down on their open weights strategy. The US model monster continue to burn they way to decimal point improvements in benchmarks with ever decreasing returns, the Chinese models from Z.Labs through to Deepseek and Alibaba's Qwen, are showing close to foundation quality for a fraction of the cost.

The Trump administration hates losing, but even they were forced to back down from the idea of limiting access to Chinese models because they are so popular with US startups. Beyond that, it would be fascinating to see how many corporates even are moving some workloads to the Chinese models.

One thing is for sure however, that we don't need to see the data to feel the impact of their popularity with the US. I see it with usage alone. Most Chinese models running on the global inference platforms perform beautifully during European working hours, and especially in the morning, but as the Americans awake, you can see it. Latency increases, tok/s drops and the occasional dropout or stall is seen, which never happens outside of US working hours.

It is clear, in numbers as in experience that US innovators are voting with the feet (and wallets) - models are commoditising and the Chinese open weights are winning...