The user asked whether higher-quantized versions of the same model perform better with long contexts, citing Qwen Flash Next at q6 and q8. They said they had expected q6 to be lossless and wondered whether context length could amplify differences between the two versions.