Gemma 4 26B A4B Instruct is a mid-sized open-weight model from Google's Gemma family, designed for instruction-following tasks with a lean active-parameter footprint thanks to its mixture-of-experts architecture. It handles general reasoning and conversational tasks competently, balancing capability with efficiency — useful when you want a capable model you can run yourself without committing to the full parameter count.