ABSURD. Qwen 3.8 27B is cruising locally on an RTX 4060 with just 8GB of VRAM.
That’s a 64k context window courtesy of Unsloth’s brand-new IQ4_XS quant, and the whole thing weighs in at 14.6GB on disk. Prefill lands around 150 tok/s; decode hovers at ~5 tok/s with native MTP.
Claude Opus 5 is INSANE
Look what I’ve just built.
"decompwlj 3D" by Opus 5: interactive app of 3D graphs with three.js - 10 sequences decomposed
⬇️
Tencent's Hy4 Preview just leaked before anyone was ready, and this breakdown covers exactly what's…
Tencent's Hy4 Preview just leaked before anyone was ready, and this breakdown covers exactly what's inside it, including the benchmark numbers most coverage is going to skip past. This video breaks down TencentHunyuan newly released Hy4 Preview, a 770 billion parameter mixture…
You have reached the end of the archive
All of Claude Opus 5