Qwen 3.8 Flash Next made a voxel Middle‑earth zoom chain + a 15 s orbit clip.
Every frame came off one RTX 3090. How the agent does it without me touching a keyboard. The interesting constraint: my agent's brain (Qwen3.8‑Flash‑Next IQ3, llama.cpp, ~21 GB) is resident on the
How to get started with open AI models (with
I have recorded a simple video for all of you who ask us. 1. How to enter NaN 2. What NaN is like inside 3. How to configure your api-key Pum, pum, pum! Quick, simple and for the whole family.
A new 1-day involving ImageMagick + LibRaw.
I was able to take it to RCE using just Qwen 3.8. No frontier model. No huge research setup. That is the part I find interesting. Once a patch/fix is public, the attacker already has a very useful starting point. Now add AI to:
Last post I showed you how to get over 100 tok/s on two DGX Sparks running Qwen 3.8 Flash.
Today I confirmed the a 2x speed-up is coming for Local AI. The answer is TensorFold. I'll tell you how I ran the smartest model on desk AI using this new invention. Same two Sparks. Same
You have reached the end of the archive
All of qwen38