So, slower and worse than Gemini 3.7 (high)?

I don't get what the issue is. Chinese labs fully publish how they make great models. Architecture is known, training methods are often very open. Why Europe just can't copy what they do?