Qwen3.6-27B on RTX 3090: tested 12 GGUF quants across HumanEval+, MBPP+, perplexity, throughput and needle-in-haystack. First-timer results.

Article automatically generated from technical news.

## Disclaimer first I'm new to local LLMs — this was my first serious attempt at benchmarking and I'm posting the results in the hope they're useful to others, not because I'm claiming any expertise. I almost certainly made methodology choices that someone more experienced would do differently. Spe