UNREAL. Qwen 3.8 27B is now happily running on an RTX 4060 locally with only 8GB of VRAM.
You’re looking at a 64k context window thanks to Unsloth’s fresh IQ4_XS quant, and it’s just 14.6GB on disk. Prefill is around 150 tok/s, decode sits at ~5 tok/s using native MTP. Only
Used Hermes and Opus 5 (it's not my favorite, but I wanted to see its capabilities in 3D)
Note: I've noticed that Hermes asked Opus automatically to stop at certain amount of time so this might be the reason of bugs (Opus said to Hermes, you didn't allow to fix the bugs 😂) this
Opus 5 + higgsfield just outworked a 5-PERSON team on a $35K project.
In one weekend. For $20. Here's the proof. not "will outwork." already did. last weekend. while their team was scheduling a kickoff call - the site was already live. here's how that weekend actually
Opus 5 vs Grok 4.6, same prompt: build a working 3D steam engine with three.js.
You have reached the end of the archive
All of Claude Opus 5