TinyLlama delivered the strongest responsiveness on the Pi, making it the most usable option for lightweight local inference. DeepSeek-R1 produced richer reasoning output but incurred much longer ...