Qwen3.8-27B vs Qwen3.6-27B writing ray-tracers in BASIC

A developer compared Qwen3.8-27B and Qwen3.6-27B in their ability to write BASIC ray-tracers. Using an agentic harness and a BASIC-to-JS transpiler, the models generated programs to render metallic spheres over a checkerboard plane with a deep blue sky, employing the Cook-Torrance model. While Qwen3.6 required user input to correct errors, Qwen3.8 consistently achieved good results autonomously, leading to a positive assessment of Qwen3.8's performance.