A compact multimodal model that handles both text and image inputs, quantized to 8-bit weights and activations for reduced memory footprint without abandoning capability. The w8a8 quantization means it runs leaner than its full-precision counterpart, making it practical on constrained hardware. It inherits Qwen3.5's general reasoning and vision understanding, though quantization introduces minor precision trade-offs compared to the base model.