Same prompt on Qwen 3.8 Q4_K_XL reasoning low 13 mins on 2x3090 24gb.
It took 36.9k tokens.
Qwen 3.8-Flash DESTROYED Opus and Deepseek
It's very strong in frontend and coding it's even sometimes compared to Fable 5 in benchmarks, it destroys Deepseek V4 Flash and Opus 4.6 while being ~2.5 times cheaper than all its competitors with Atomic Agent, this model
Qwen 3.8 27B 4bit To Do List App - Zcode run, serving over mtplx
No voxel pagoda prompt can show differences such clearly as 2 simple apps The quality difference on the UI between Qwen and Ornith is clealy visible. And truthfully Qwen made an actual real production grade app
Qwen 3.8 Flash Next NVFP4 on 2xRTX 6000 Pro; one-shotted.
The interesting part: The video would drop frames on my laptop, so Qwen went to my server, rendered it in headless Chrome, and returned the full video!
You have reached the end of the archive
All of qwen38