The capabilities of various large models are at least converging in terms of scores more and more; as long as there are continually open-source base models, it’s only a matter of time before everyone catches up—it's basically a foregone conclusion.

Remember what was said at the start of the year: that 3 would be the worst model this year? As of now, it really can’t get onto the table anymore.
But in the end, when large models fight to the finish, they’re still competing for channels—GLM’s government and enterprise orders, Qwen’s iOS domestic (China) version. The race among AI large models is about to accelerate into the second half; losing channels means losing everything.