Google's move to fit large models into 1GB of memory, making AI run smoothly on mobile devices, finally eliminates lag, and developers are ecstatic.

CryptoWorld News reports that Google has released the ultra-lightweight Gemma 4 model, with mobile local runtime memory dropping below 1GB for the first time. The model uses quantization compression technology, reducing numerical precision to shrink the model size while maintaining a high level of intelligence. Google has also optimized it for mobile chips to ensure smooth performance. The new model weights have been open-sourced on Hugging Face; individual users can download and run them via Ollama and LM Studio, and mobile and web developers can also deploy quickly through supported engines.