Google has officially launched Gemma 4, a new family of fully open-source AI models licensed under Apache 2.0, marking a significant expansion in its AI offerings. The models, which are designed for deployment from data centers to billions of Android smartphones and laptops, support advanced multimodal inputs including text, image, audio, and video, according to Mashable and VentureBeat.
Gemma 4 includes a range of model sizes, with a flagship version boasting 31 billion parameters and smaller versions optimized for offline mobile use. This versatility allows for enhanced privacy and developer flexibility, as users can run the AI locally on their devices instead of relying solely on cloud processing, The Register and Times of India report.
The new models also support over 140 languages and feature agentic workflows and improved reasoning capabilities, positioning Gemma 4 as a contender in both general AI tasks and coding-specific applications. Google DeepMind’s adoption of the permissive Apache 2.0 license is seen as a strategic move to attract enterprise developers and to compete with other open-weight AI ecosystems, according to VentureBeat and Ars Technica.
Gemma 4’s open licensing and multimodal functionality aim to revive Google’s footprint in the open-source AI community, where it had previously seen diminished visibility. Analysts highlight this launch as a key step for Google in the US open-weight model race, expanding the accessibility of powerful AI tools to developers at all levels, as reported by Decrypt.
Looking ahead, industry watchers will closely monitor how widely Gemma 4 is adopted across devices and enterprises, and whether the Apache 2.0 licensing encourages wider community contributions and usage. The ability to run these sophisticated models locally while maintaining privacy could reshape expectations around AI deployment on consumer and edge devices.
Kaggle
Gemma 4
Ollama
Google
Hugging Face
DeepMind
