Alibaba, a major tech company, has just released a new version of its AI model, and it comes with a cool change in how they’re building it. They’ve decided to train their AI in two separate ways to make it even better. Let’s break down what this means for you.
A New Strategy for Better Quality
Previously, Alibaba was working on a “hybrid thinking mode” for their AI. Think of this like trying to teach a student to be both a great listener and a great problem-solver all at once.
Now, they’ve changed their minds. After talking with the tech community, they’ve decided to train two different types of models separately: “Instruct” models and “Thinking” models.
- An Instruct model is trained to be really good at following specific commands and instructions you give it.
- A Thinking model (which they are still working on) will likely be designed for more complex, deep problem-solving.
By separating them, they believe they can make each model achieve the “best quality possible” in its specific area.
Meet the New Model: Qwen3-235B-A22B-Instruct-2507
The first model released under this new strategy is called Qwen3-235B-A22B-Instruct-2507. That’s a long name, but the important part is “Instruct.” This new AI is designed to be better than their last one, with what they call “strong overall abilities.”
According to Alibaba, this new version is:
- Smarter
- Knows moreÂ
- Can do more things
- Works better on agent tasks (meaning it’s better at completing multi-step tasks for you)
How Can You Try It?
The best part is that you can try it out yourself!
- For a simple chat: You can go to chat.qwen.ai.
- For developers: If you’re more tech-savvy, you can find the model on platforms like Hugging Face and ModelScope.
Alibaba encourages users to test it and see how it works. They’ve also mentioned that while this is a nice improvement, it’s just a “small update.” They hinted that “Bigger things are coming soon!”, so we can expect more exciting news from them in the future.