With the release of Qwen3-Coder, I wanted to do a practical test of its "agentic coding" capabilities by building several real-world applications entirely from text prompts.

The linked video is a full walkthrough of that process. In it, I cover:

The Platform: A walkthrough of the Alibaba Cloud Model Studio, which provides free access and a large token allowance for testing the Qwen3 models without a local setup.

The Projects: I prompted the model to build five distinct applications from scratch: a 3D car racing game, a plane fighter, a zombie dragon FPS, an AI image generator website, and a web-based music sequencer.

The Results: For each project, I show the full code output (HTML/CSS/JS) and then immediately paste it into CodePen to see if it runs and how well it performs.

A few takeaways: The model's ability to generate the complete code for complex, interactive apps was genuinely impressive. The music sequencer worked nearly perfectly on the first try, while the 3D games required more specific prompting to get the controls right (a good lesson in prompt engineering).

Curious to hear what you all think about its performance and the direction these kinds of coding tools are heading. Happy to answer any questions.