原帖内容

The Big Tetris Battle 🏆 I've asked GLM-5.3, Grok 4.6, DeepSeek v4 Flash, Kimi K3, Qwen3.8-Max, and Qwen3.8-27B to create the most impressive Tetris game they can in a single shot. To keep things fair, I used ZCode for GLM-5.3, Grok Build for Grok 4.6, DeepSeek Harness for DeepSeek v4 Flash 0731, Kimi Code for Kimi K3, and Qwen Code for Qwen3.8-Max and Qwen3.8-27B. All prompts were with highest reasoning/effort. They've all built a working game without any errors. All models included sound, and most also included background music, except for Qwen3.8-27 and Kimi K3, which had no music. Results are live here: https://miaai-lab.github.io/Tetris-Battle-DS4F-Grok46-GLM-5.3-Qwen38-27b/ IMO, the best are of DS4F, GLM-5.3 & Grok 4.6. Interestingly, Grok 4.6 was the only one that didn't use the typical purple color associated with “AI slop” and actually felt more unique. Kimi K3 was also the only one that made the game full-screen. Tokens burned by each model - GLM-5.3 → 156k - Grok 4.6 → 178k - Qwen3.8-27B → 170k - Qwen3.8-Max → 179k - DeepSeek v4 Flash → 266k(!) - Kimi K3 → 106k The biggest takeaway for me: We've reached a point where all of these models are excellent. Qwen3.8-27B is really impressive and can absolutely be compared to much larger frontier models. Github repo: https://github.com/MiaAI-Lab/Tetris-Battle-DS4F-Grok46-GLM-5.3-Qwen38-27b Demo videos & full prompt in posts below 👇