Gemma 4

Hitoku Draft is an open-source, voice-first AI assistant that runs locally, offering transcription, voice editing, and context-aware features by analyzing the user's screen and documents. It supports text generation models like Gemma 4 and Qwen 3.5, and the creator is seeking feedback while developing a cross-platform version.

Google has developed a new technology to speed up local AI processing, making it up to 3 times faster without requiring new hardware or cloud services. This innovation uses Multi-Token Prediction drafters to improve Gemma 4 performance. The breakthrough enables faster AI processing on existing devices.

Google's Gemma 4 AI models have received a 3x speed boost through Multi-Token Prediction, allowing for faster generation of tokens. The latest models are built on the same technology as Google's Gemini AI and can run locally on consumer hardware. This development enables users to work with AI on their own devices without relying on cloud-based systems.

Google introduces Accelerating Gemma 4, a new technology for faster inference with multi-token prediction. The innovation aims to improve developers' tools. The article is available on Google's blog.

Google's Gemma 4 AI model now runs natively on iPhones with full offline inference capabilities, marking a significant advancement in mobile AI technology. This development highlights the growing integration of advanced AI models into consumer devices.

The article discusses running Gemma 4 as a local model using Codex CLI. It highlights the technical implementation and availability of the model through a CLI tool.