It has been a long time since we talked about AI, and the recent progress of Chinese models is indeed exciting.
The impact of this matter could be significant because:
The United States is still locking the strongest models in safes and selling tickets per token.
China has already begun to throw the capabilities close to the strongest models, along with the keys, onto the internet.
After Kimi K3 was released, it temporarily surpassed Fable 5 on the front-end programming leaderboard, shooting straight to the top.
How hot is it?
Within a few days of opening subscriptions, users overloaded the servers.
The model hasn't dominated the world yet, but the data center is already asking for a break.
Next, Alibaba released Qwen3.8-Max-Preview.
With 2.4 trillion parameters, it supports text, images, videos, and documents.
Alibaba's own positioning is quite straightforward:
Second only to Fable 5.
Of course, we must pour a little cold water on this.
Kimi K3 surpassing Fable 5 mainly occurs in the front-end programming category and does not represent a comprehensive capability lead.
Qwen3.8 being "second only to Fable 5" is currently also Alibaba's own claim, lacking large-scale independent evaluation.
But what's really important isn’t about who scored 20 points more on a particular ranking.
It's that Chinese models are starting to sit at the main table of cutting-edge models and are preparing to open weights.
What was the AI competition between China and the US like before?
The United States was responsible for inventing the latest sports cars.
China was responsible for catching up, lowering prices, and then making a viable alternative.
Now, the situation has changed.
Kimi and Qwen have begun to bring their cars onto the same track and then turn to tell the whole world:
The design blueprints will be shared with you in a few days.
This is what truly gives American AI companies a headache.
Even if Fable 5 remains the strongest, as long as open-source models can reach most of its capabilities, a large number of global developers will not care about the final point difference.
They care more about:
Can they deploy it themselves?
Can they modify it?
Will prices suddenly rise?
Will the API go offline one day?
Do they need to share their own data with others?
The American large models want to sell "intellectual luxury goods."
The Chinese large models are competing for "global intellectual infrastructure."
One relies on leading to charge high ticket prices.
The other relies on open-sourcing to allow everyone to enter the construction site.
Of course, opening weights does not equal complete open source.
Training data, training methods, and complete technical details may still not be disclosed; this kind of trillion-parameter model cannot simply run on home computers, as it still requires expensive computing power behind it.
The overload of Kimi's servers also indicates that the capabilities of Chinese models have caught up, but their service capabilities and computing power reserves have not fully kept pace.
But the trend is already very clear:
The gap between Chinese models and the strongest American models is narrowing from several generations to several months, or even weeks.
The United States still has the most advanced chips, the strongest laboratories, and the most mature commercial ecosystems.
But China has found a very aggressive strategy:
If it can’t temporarily dominate the strongest intelligence, then turn the near-strongest intelligence into public resources.
Fable 5 may still be on the throne.
But Kimi and Qwen have already started to dismantle the floorboards under the throne.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
