Developer Successfully Builds Free Web Game Using Qwen 3.8 Flash Next
A developer successfully built and shipped a free-to-play web game entirely using the local Qwen 3.8 Flash Next model for coding assistance without relying on frontier models. The project was developed using TypeScript and Three.js. This demonstrates the practical viability of using smaller, local language models for end-to-end application development and coding tasks. It highlights how local AI tools can autonomously handle substantial coding workloads when paired with proper human oversight and debugging tools. The project involved processing over 20 million output tokens and 31,883 tool calls using an Unsloth quantization of Qwen 3.8 Flash Next running on an AMD Strix Halo machine. The author emphasizes that while the model excelled at coding, it required strict scope control and custom human logging tools to prevent shortcuts and logic errors.
## BACKGROUND
Qwen is a family of open-weight large language models developed by Alibaba, while Unsloth provides optimized quantization techniques to run models efficiently on consumer hardware. Three.js is a popular JavaScript library used to create and display animated 3D computer graphics in web browsers.