Qwopus tops local flight sim benchmark

This is not a rigorous leaderboard, but it is a useful artifact test because it measures whether models can ship something playable, not just pass canned coding evals.

• –Qwopus winning with fewer prompts and real flight physics is a strong signal for distillation quality on messy creative coding tasks.
• –The three Qwen3.6 35B quant variants behaving differently makes quantization source a practical evaluation dimension, not just a packaging detail.
• –Qwen Coder Next 80B producing the most code but one of the weakest games is a reminder that bigger code output can hide worse product judgment.
• –The wildcard plane choice worked as a lightweight creativity probe, exposing differences that standard benchmark prompts usually miss.