GLM-5 Is a local GOAT

A developer with decades of experience tested the new GLM-5-744B model on high-end local hardware. In just three prompts, the model generated a fully playable, GPU-accelerated Flappy Bird clone with retro styling and even provided an AWS cost assessment for viral scaling. While output was slow at 16.5 tokens/second, it reportedly outperformed other leading open models like Qwen3-Next-80B and GPT-OSS-120B in this real-world task.