Last Week, Lao Zhang Panicked
Last week, Lao Zhang, who runs a Xiaohongshu account, asked why his AI writing tool had suddenly “gotten dumber.” We found that the platform had switched its underlying model—right after Alibaba released a much larger Qwen3-series model with 2.4 trillion total parameters.
What It Is—and Who’s Using It
In plain terms, Alibaba’s Qwen team released a giant AI model with 2.4 trillion total parameters, 95 billion of them active at a time. It can write marketing copy, analyze information, and answer questions more capably than the previous generation.
Why does that matter to anyone using Tongyi Qianwen, Kimi, Wenxin Yiyan, or similar tools? They all run on this kind of underlying model. When a new one improves, the old model can suddenly feel less capable—not because the tool broke, but because its engine changed.
Xiaolin, an e-commerce copywriter I work with, tested the Tongyi Qianwen web app last week. It produced sales copy that felt more human than before. She had no idea a model of this scale was behind it.
What It Costs to Try Today
Money: The web app is free. API access, which lets software call the model automatically, costs roughly a few fen—fractions of a yuan—per paragraph. Self-hosting isn’t realistic; this model is so large that it would take dozens of GPUs to run.
Time: Zero minutes for the web app. One to two hours to connect the API. Zero minutes to wait for a platform to integrate the upgrade automatically.
Technical barrier: None for the web app. The API requires understanding an API key—a platform-issued password that proves the request belongs to you and prevents others from burning your quota.
First step: A practical way to start is opening tongyi.aliyun.com, signing in with an Alipay or Taobao account, selecting Tongyi Qianwen, and starting a chat.
Advice by Stage
Just starting out, with no clients yet: There’s no need to overthink it or rush into testing. The tools already in use will likely upgrade to stronger models on their own.
With one or two clients: I’d run the same passage through Tongyi Qianwen and compare it with the current tool. If the result is clearly better, switching is worth considering. I once made the same mistake: when an AI tool got worse, I assumed the tool itself was broken. In reality, the underlying model had changed.
Scaling up: If monthly AI costs exceed ¥200, API access may be worth considering for more cost control. Otherwise, waiting for the platform to upgrade is the better use of time; chasing every new release adds work without proving value.