Google takes a decisive step in the world of artificial intelligence with Gemma 4, its new open source model that combines power and flexibility, ready to run on both mobile devices and laptops without limits. With a fully open license, the company opens the doors to developers and users, offering total control over data, infrastructure, and models, without sacrificing privacy or security.
Gemma 4 is available on Google AI Studio and can also be downloaded via third-party platforms such as Hugging Face, Kaggle, and Ollama.
THE LAUNCH OF GEMMA 4
Google has announced the release of Gemma 4, the latest version of its open artificial intelligence model. Unlike many cutting-edge models, Gemma 4 is distributed as fully open source software under the Apache 2.0 license.
Open models can be run directly on users’ devices and, according to the company, Gemma 4 is capable of running on “billions of Android devices” and on some laptop GPUs.
“This open source license provides a foundation for complete flexibility for developers and digital sovereignty; giving you full control over data, infrastructure, and models,” reads a post on Google’s blog. “It allows you to freely build and securely deploy in any environment, whether on-premises or in the cloud.”
THE RELATIONSHIP WITH GEMINI
The new model sits alongside Gemini, already widely used in the company’s products. Gemini represents Google’s proprietary artificial intelligence platform and is integrated into services such as Search, Gmail, Docs, and Cloud.
Gemma, while based on the same technology developed by Google DeepMind for Gemini 3, follows a different approach. It is, in fact, a large language model designed to be accessible and usable locally, without an internet connection.
Google describes Gemma 4 as its most “powerful” open model to date.
ADVANTAGES OF THE OPEN MODEL
As mentioned, Gemma 4 stands out for its open nature, which allows users to access both the code and the training data. The model can be downloaded for free and run on local hardware, with no subscription required.
This setup ensures greater privacy since chats, uploaded files, and responses are not shared with third parties. At the same time, developers can integrate artificial intelligence into their applications while avoiding recurring costs.
TECHNICAL FEATURES
According to Google, Gemma 4 introduces advanced reasoning capabilities, including multi-step planning and complex logic. The company also reports “significant improvements in math benchmarks and in the ability to follow instructions that require it.”
The model supports agentic workflows and local programming assistance tools. It is also capable of processing audio and video, handling tasks such as speech recognition and interpreting visual content, including charts.
Gemma 4 is available in four variants, with 2, 4, 26, and 31 billion parameters. It was trained on over 140 languages and offers a context window of up to 256,000 tokens, while the lighter versions stop at 128,000.
FROM OPEN-WEIGHT TO OPEN SOURCE
In previous versions, notes Mashable, Gemma was defined as open-weight, meaning the datasets were public, but usage remained bound by terms set by Google. Users could modify the model locally but had to comply with strict rules on usage and distribution.
With Gemma 4, this limitation is overcome. The model is now both open and open source and is distributed under the Apache 2.0 license, which allows use, modification, and redistribution even for commercial purposes, without the obligation to pay royalties. The only requirement is attribution, along with distribution of the license.
