Tag
This article reports benchmark results showing that quantization has little effect on knowledge benchmarks (GPQA) but significantly degrades agentic performance (Terminal-Bench 2) for Qwen 3.6 models, with further observations on timeout settings and run variability.