I think it's actually due to the fact that Claude isn't available on China, so they wouldn't be able to (legally) replicate how they evaluated the other LLMs (assuming that they didn't just use the numbers reported by each model provider)