Google Releases Open Gemma 4 AI Models for Local Device Use
Similar Articles
Google Chrome Gains On-Device AI Features for Writing, Organization, and Security
Google Home Update Brings Gemini 3.1 AI to Smart Speakers
EU Launches Investigation Into Google's AI Integration on Android
OpenAI Unveils Enhanced Image Engine for ChatGPT
OpenAI Rolls Out Less Restricted GPT-5.5-Cyber to Vetted Security Defenders
Google has launched its Gemma 4 open models, which are tuned to run on local hardware. The release includes a more permissive Apache 2.0 license and experimental speed-boosting technology. This allows users to run AI on their own devices instead of relying on cloud systems.
Facts First
- Google has launched its Gemma 4 open models, which are built on the same underlying technology as its Gemini AI.
- The models are tuned to run locally on consumer hardware, allowing users to run AI on their own devices.
- Google changed the license for Gemma 4 to Apache 2.0, a more permissive license than used for previous releases.
- The release includes experimental Multi-Token Prediction (MTP) drafters that use speculative decoding to speed up generation.
- Quantization allows the models to run on a consumer GPU, while a single high-power AI accelerator can run the largest model at full precision.
What Happened
Google launched its Gemma 4 open models in the spring. The models are built on the same underlying technology as Google's Gemini AI but are tuned to run locally. Google also released experimental Multi-Token Prediction (MTP) drafters for Gemma to speed up generation. The company changed the license for Gemma 4 to Apache 2.0, which is more permissive than the custom license used for previous releases.
Why this Matters to You
You can now run capable AI models directly on your own computer or device, which may give you more control over your data and privacy compared to using cloud-based AI systems. This could make AI-powered tools more accessible and responsive for personal or professional use without an internet connection. The more permissive license may also encourage wider development and integration of these models into other software you use.
What's Next
The experimental MTP drafters could lead to faster and more efficient AI processing on local devices if the technology is refined and widely adopted. The shift to a more permissive Apache 2.0 license may encourage broader community and commercial adoption of the Gemma models. Developers and companies are likely to begin integrating these local AI capabilities into a wider range of applications.