Gemma 4, Hugging Face’s latest multimodal model, now runs on-device. The 4‑billion‑parameter model supports image, text, and audio inputs, delivering inference speeds up to 10× faster than its predecessor. Developers can embed richer AI directly into mobile apps without cloud latency. Gemma 4’s open‑source license encourages community contributions. It runs on modest GPUs, making it accessible.