Ornith-1.5-35B-A3B-oQ8e-mtp in action with oMLX v0.6.3rc1 Video normal speed!
🚀 Model used here: Tomorrow I'll compare results vs Qwen 3.8 27B!
Qwen 3.8 Max (xhigh)
495 tool calls in 467 steps, 139 064 532 tokens • cool minimal UI with blurring panels • good taste and variety on default portraits • throws execute from the preview location • wall texture with ugly seams • no import-to-wall • dead trampoline
Qwen 3.8 27b 📍 Alibaba_Qwen
IQ2_XXS by UnslothAI (Local model) running on a CPU same test ... MTP off: avg speed 3.43 Time: 41 mins Rate: stunning design 😍 reply button unresponsive, thinking as medium more tokens yet bette, low was found to be worse! 221 net lines of code
Four models. Three one-shot visual builds.
No retries. No second chances. GPT 5.6, Qwen 3.8, Grok 4.6 and Opus 5 received the exact same three prompts: a premium product page, a playable browser game, an animated rainy intersection. Every model produced something that ran. The
You have reached the end of the archive
All of qwen38