Google has unveiled its new open AI model, Gemma 4, which is capable of running on various devices, including Android smartphones. This move marks a significant shift from proprietary models like Gemini or OpenAI’s GPT 5.4, as Gemma 4 is designed to be open-source, providing developers with enhanced freedom and functionality.

The announcement was made by Demis Hassabis, CEO of Google DeepMind, via a post on X. Hassabis asserted that the four variations of Gemma 4 represent the finest open models in the world corresponding to their sizes. This accolade reflects the commitment Google has to advancing AI technology and fostering an open development environment.

Notably, the open-source nature of Gemma 4 allows users to download and run it locally, which also aids in maintaining data privacy, as the model does not transmit information back to Google or any third party. This development is expected to resonate well with users prioritising data security.

Capabilities of Gemma 4

Gemma 4 comprises four sizes tailored for diverse applications, particularly in advanced reasoning and agent-based workflows. These include the Effective 2 Billion (E2B) and Effective 4 Billion (E4B) models suitable for mobile and edge devices; a 26 Billion Mixture of Experts (MoE) model aimed at low-latency tasks; and a high-performing 31 Billion Dense model. The latter currently ranks as the third-best globally according to the Arena AI leaderboard, surpassing models that are significantly larger in size.

According to Google, the advancements incorporated into Gemma 4 facilitate improved capabilities in complex logic and multi-step planning. The agentic workflows allow users to designate tasks to AI agents, enabling the potential formation of a team of AI assistants capable of carrying out various jobs, much like the functionality seen in projects such as OpenClaw.

This innovative model can process images, videos, and audio inputs for speech recognition, expanding its usability. Additionally, the enhanced context window capabilities allow models to handle extensive documents or code repositories, with the edge models managing 128,000 tokens and larger models accommodating up to 256,000 tokens. Supporting over 140 languages broadens the accessibility of the model to a global audience.

Availability and Community Impact of Gemma 4

Gemma 4’s accessibility is bolstered by its open-source release under the Apache 2.0 licence, granting developers the flexibility to utilise, modify, and deploy the models as they see fit. This release follows the previous generation’s success, where over 400 million downloads fostered a community with more than 100,000 variations of the earlier models.

In contrast to earlier versions, which were open-weight rather than fully open-source, Gemma 4’s release means that users are no longer bound by stringent terms that limited how they could operate the models. This shift could substantially influence the development of AI applications, empowering users to innovate without restrictions.

Gemma 4 can be accessed through Google AI Studio and is downloadable from various platforms including Hugging Face, Kaggle, and Ollama. This broad availability underscores Google’s intention to democratise access to advanced AI technologies and encourage widespread utilisation in diverse fields.

Long or Short, get news the way you like. No ads. No redirections. Download Newspin and Stay Alert, The CSR Journal Mobile app, for fast, crisp, clean updates!

App Store –  https://apps.apple.com/in/app/newspin/id6746449540 

Google Play Store – https://play.google.com/store/apps/details?id=com.inventifweb.newspin&pcampaignid=web_share