Qwen 3.8 Flash Next (125B-A6B) now runs natively in mlx-serve on Apple Silicon.
No Python, Zig + Metal. (Preview Release) M4 Max, 4-bit pack (~75 GB resident): - ~70 tok/s serial, ~98 tok/s with MTP when tested using `npx llmprobe` - prefill ~700 tok/s, sparse attention past 2k
Qwen 3.8 27b with your exact prompt via Hermes.
The darker hue is the first attempt/pass The more vibrant hue is the second pass mentioning the sharp lines
The quality of Qwen 3.8 flash is insane.
Look at the detail- this was literally a one shot 2-3 line prompt - I can give it below. It has individual satellites beautifully modeled, the globe looks amazing with light and dark, the glow the polish the smoothness, are you kidding me?
Qwen 3.8 27b surprises again.
I recently found an interesting video on X that showcased the capabilities of p5.js. I decided to try building something different, but using a local Qwen 3.8 27b. After an hour of tweaking, the result was unsatisfactory. Each iteration was
You have reached the end of the archive
All of qwen38