A compact multimodal model from the Qwen3.8 family, published by autotrust in a heavily quantized FP4 format. The FP4 precision and 'Flash' designation suggest this is optimized for speed and reduced memory footprint, likely at some cost to output fidelity compared to higher-precision variants. It accepts both text and image inputs, making it capable of vision-language tasks within its constrained format.