Ali's a good player. Put the big model on the phone. There's always one for you.

Just today, Ali launched a new lightweight small model in the Qwen3.5 series, with parameters of 0.8B, 2B, 4B and 9B versions. This is about putting AI power directly into your phone and computer!

The age of small models.

We used to talk about the big AI model, and the first reaction was "burn cards" -- there were no 3090, 4090 cards, and they didn't move. But this time, Ali's Qwen3.5 small model series completely subverted that perception.

What is the concept of 0.8B? The equivalent of 800 million parameters. On this scale, you can even run on some smart home equipment. 2B is 2 billion, 4B is 4 billion and 9B is 9 billion. The 9B version is already quite powerful and close to the level of some of the earlier medium and large models.

Why push small models?

The large model is strong, but there are three fatal questions:

• High deployment costs: Tens of thousands of GPUs are required to train a large model at a cost of hundreds of millions of dollars
• Reaction slow: a 70B model that just loads to visibility takes seconds
• Can't deploy locally: companies want to use AI for privatization deployment? It's impossible.

Small models have solved these problems perfectly. You can run on your computer, you can run on your phone, you can even run on a strawberry pie. Data are not uploaded and are fully private.

Developer ecstasy

For the developers, this wave is really good. Ali's operation is equivalent to an AI ticket to a global developer.

You used to want to be an AI app. Model deployment alone can get rid of a bunch of people. Now, 0.8B models can run by a computer, 2B, 4B doesn't require high graphic cards, and 9B versions of the ordinary game book can be carried.

What does that mean? This means that the threshold for AI application development has been further lowered. And then ordinary people can help themselves with AI help, make their own local smart clothes, without looking at the big factory.

Nationally A.I.

From Qwen2.5 to Qwen3.5, Ali is an interesting move. Before that, major manufacturers were rolling large models of hundreds of billions of parameters, and Ali was now on the contrary, starting to roll small models.

And behind this is a new trend in AI's development: not just the cloud brain, but the local brain. In the future, your cell phone, watch, car, you'll probably have a local AI model, you'll have simple problems, you'll have complicated problems and you'll be in the clouds.

The release of the Ali wave small model marked another step forward in the deployment of the National Production AI model.

Did you take a good look at the small model? Welcome to the comment section!