alt Hacker NewsChinese AI models have lagged the US frontier by 7 months on average since 2023
Comments
On the flagships, maybe. But on open models (and especially the small ones), China is kicking ass.
They have been catching up pretty fast though. Since US have banned Nvidia chips export upto a certain extent to them they are coming up with more optimized training at least which is why US should be wary of them. China does more with less
State of the art 7 months ago is good enough for a lot of use cases.
7 months on average and decreasing. As the rate of progress in models slows down the head start of any entity in the field will slowly reduce until there is parity.
Nice. So ~ half a year and we get open-weights Opus 4.5 capability? I'm all for that!
Question about English for natives: "[...] have lagged behind [..]" would be the grammatically correct version of the heading, I think. Or is "to lag" without "behind" actually a correct use? Is it merely headline-speak, news-speak, to make headlines shorter and convey more information in fewer words?
I could not find GLM on there, it is one of the best open weight Chinese models now, probably not at the top but seems only a few months behind.
that's because they are distilling the frontier models
And how much have they spent compared to US companies? What of carbon footprint?
so there has been quite a few instances of extensions logging AI chatbot conversations, I wonder if this is related to them training on that data given the accusations of deepseek 'stealing' chatgpt.
So basically the amount of time it takes for them to scrape existing US models and then train on that data. China doesnt have any intellectual property, they are just stealing
They use the ECI - EpochAI Capability Index.
Measured by the DCI the Chinese AI models are about 1.5 years ahead of US models.
DCI = Dust42 Capability Index: MBP Max 64GB, Qwen3-80B MLX 4bit quant, 40 tokens per second. It is not on Claude Opus level but very, very useful if you have no internet, i.e. on a flight. And occasionally it surpasses even Opus by far and large. Opus is a pain in the neck once the coding task at hand surpasses its capabilities. Qwen3 is much better to guide to get step by step to a solution.