Google launches Gemma 4 AI models for data centres and smartphones: What CEO Sundar Pichai and Demis Hassabis have to say

Google has unveiled Gemma 4, the newest and most advanced version of its open AI model series, designed to bring powerful artificial intelligence capabilities to both large-scale data centers and everyday devices like smartphones. The company highlighted that the release marks a major step toward making high-performance AI more accessible to developers worldwide. Since the launch of the earlier Gemma models, downloads have surpassed 400 million, with a growing ecosystem of over 100,000 community-built variations.

According to Sundar Pichai, the latest models deliver exceptional intelligence relative to their size, while Demis Hassabis described Gemma 4 as among the top-performing open models available in their respective categories.

Gemma 4 is released under the Apache 2.0 license, allowing developers to freely use, customize, and deploy the models for various applications.

Multiple Model Variants for Different Needs

The Gemma 4 lineup includes four versions, each tailored for specific use cases:

• E2B (Effective 2B parameters) for mobile and IoT devices
• E4B (Effective 4B parameters) optimized for edge computing
• 26B Mixture of Experts model for balanced performance and efficiency
• 31B dense model, the most powerful version, currently ranked among the top open AI models globally

Notably, the 31B model reportedly outperforms systems significantly larger than itself, showcasing improved efficiency.

Enhanced Capabilities

Gemma 4 goes beyond basic conversational AI by offering advanced functionalities such as:

• Improved reasoning and multi-step problem-solving
• Support for building autonomous AI agents with tool and API integration
• Offline code generation, enabling developers to use it as a private coding assistant
• Native processing of images, video, and audio inputs
• Extended context handling, supporting large volumes of data in a single interaction
• Training across more than 140 languages, making it highly adaptable for global use
• Optimized for Smartphones and Edge Devices

One of the most notable advancements is the ability of smaller Gemma 4 models to run directly on devices without requiring cloud connectivity. Developed in collaboration with hardware partners like Qualcomm and MediaTek, these models deliver fast performance with minimal latency on smartphones, embedded systems, and edge hardware.

Google also noted that Gemma 4 is built using the same foundational research that powers its flagship AI systems, positioning it as a strong contender in the open AI landscape.

Voice Of Osiz

At Osiz, we view Google’s launch of Gemma 4 as a major leap in democratizing advanced AI capabilities across industries. The ability to run powerful models on both data centers and edge devices highlights a new era of accessible and efficient AI deployment. Insights from leaders like Sundar Pichai and Demis Hassabis reinforce the growing importance of scalable, high-performance open models. Features such as offline functionality, multilingual support, and agentic workflows reflect the evolving needs of modern businesses. This development empowers organizations to build smarter, faster, and more secure AI-driven solutions. At Osiz, we believe such innovations will accelerate enterprise adoption and reshape digital transformation strategies. We remain committed to delivering cutting-edge AI solutions aligned with these emerging technological advancements.

Source: The Times Of India