Sad. Qwen3.8 27B IQ3_XXS isn't living up to my expectations.
It looks like I'm unable to find a good local LLM for 16 GB of VRAM.
I'm starting to think there simply isn't one.
I keep repeating...
GPT-OSS-20B is a really fast model, but since it's not following instructions it's not useful at all for coding. It just skips all checks, so at the end its output doesn't even compile.
Stefan Oltmann
08/30/2026, 8:49 PM
I ended up subscribing to z.ai GLM Coding Lite for $18/month