A 28.9 million parameter language model generates text on the ESP32-S3 microcontroller, which costs about $8. The model operates entirely on the chip without server communication, displaying output on a small screen at approximately 9 tokens per second.

Google's Gemma 4 model family now runs natively on iPhones, enabling full local AI inference offline. Early benchmarks show the 31B variant of Gemma 4 performing comparably to Qwen 3.5's 27B model.