AI Coding Hype vs. Reality: New Model Fails to Deliver

A Reddit user recently shared their experience with the new Qwen 3.8 Flash Next AI model, which many people online have been praising as fast and great for coding. The user decided to test it by asking the AI to build a Windows Tetris game with keyboard controls, scoring, and a two-player mode. They ran the model on powerful hardware and compared it to an older, larger model called Qwen 3.8 27b.

The results were disappointing. The Flash Next model produced buggy code—for example, the game controls still worked when the game was paused or over. Even after four hours, it failed to create a playable game file. In contrast, the older 27b model fixed its own bugs during a second attempt and produced a working game. The user noted that while the Flash Next model generated text faster, the quality was worse.

This matters because many people are excited about new AI models that promise to write code quickly and cheaply. But as this test shows, speed and hype don't always translate to good results. For anyone hoping to use AI to build software without programming skills, this is a reality check: the technology is improving, but it's not yet reliable for complex tasks. The user's experiment highlights the importance of testing claims and not believing the hype blindly.