The horror of Qwen 3.8 unc version
Qwen 3.8 27B running locally on a single RTX 5090 is getting ridiculous.
It beat Opus 4.8 in my benchmarks and hit ~200 tok/s in some tests.
So here I must admit that I am amazed, I absolutely did not expect it.
Qwen 3.8 27B Q4 with MTP, limited to 64k context on a single 3090, with the OpenCode harness. A single prompt: "Dev a realistic view of sea water with light etc. with threeJS or other engine more
Launching Qwen 3.8 with its 27B model is like giving someone a Corvette engine for their kitchen table—useless without the right setup.
Just having the GGUF file isn't enough. LLMs need a harness with tools like web search to be truly functional.
You have reached the end of the archive
All of qwen38