Qwen/Qwen3.5-35B-A3B creates FlappyBird

A viral demonstration shows Alibaba's open-source Qwen3.5-35B-A3B model successfully building a fully-featured Flappy Bird clone from vague initial prompts in just two hours. Using the Roo extension in VSCode, the developer started by asking for "a flappybird clone in html, css and typescript" initialized with Vite. The model's initial output was so impressive that the developer continued requesting additional features, including programmatically generated sound via the Web Audio API (requiring no external files), a scrollable parallax mountain background, and an animated flock of birds that flies across the screen. The model handled these iterative requests with minimal back-and-forth, successfully interpreting descriptive guidance like "they fly off screen" and "make them fly from right to left."

The technical achievement highlights the rapid maturation of locally hostable, open-source coding models. The Qwen3.5-35B-A3B, a 35-billion-parameter model, managed a complex multi-step project involving game logic, dynamic audio synthesis, and advanced CSS animations. Notably, it "one-shotted" the creation of a sound settings panel. This performance challenges the notion that only massive, proprietary models like GPT-4 or Claude 3.5 are capable of general coding tasks. For developers, it signals a shift toward viable, private, and cost-effective AI coding assistants that can execute on creative, feature-rich projects without constant hand-holding, potentially accelerating indie game development and prototyping workflows.