Heavily quantized Qwen3.8-Flash vs Q8 Qwen3.8-27B - thoughts?

A developer is seeking opinions on the performance of two heavily quantized Qwen models: IQ3_XXS Qwen-3.8-Flash and Q8_0 Qwen3.8-27B. They currently lack complex tasks to properly evaluate the intelligence of these models and are asking the community for comparisons and experiences to determine which offers a better user experience.

This post directly compares the heavily quantized IQ3_XXS Qwen-3.8-Flash and Q8_0 Qwen3.8-27B, unlike other discussions that focus on individual model performance.