Ironing Platypus Test DeepSeek V4 Flash 0731 vs Qwen 3.8 27b
What would you do if you burned your own body?
Round 2: same setup, harder prompt — an endless procedural terrain flythrough.
FBm noise, real vertex normals, altitude colour banding, distance fog. Qwen 3.8 got it clean first try in 9 min. Qwen 3.6 shipped a broken shader, needed a second pass, and took 20 min total.
Game I built with Qwen 3.8 27b NVPF4
Same prompt, same machine: Qwen 3.6 vs Qwen 3.8.
"Build an animated 3D spiral galaxy" in three.js, run through a coding agent that feeds real shader errors and measured frame times back to the model. Two attempts each. Left is 3.6, right is 3.8.
You have reached the end of the archive
All of qwen38