Google DeepMind Unveils Gemma 4: Open Models for Multimodal AI on Various Devices

Google has released Gemma 4, a family of open models that surpass previous capabilities in multimodal intelligence, including vision input and audio processing. The new models are designed to be deployable on various devices, from high-end phones to laptops and servers, making them accessible for tasks such as text generation, coding, and reasoning.

What is Gemma 4?

Gemma 4 is a family of open models built by Google DeepMind. The new models are multimodal, handling text and image input (with audio supported on small models) and generating text output. This release includes open-weights models in both pre-trained and instruction-tuned variants. Gemma 4 features a context window of up to 256K tokens and maintains multilingual support in over 140 languages.

The models are available in four distinct sizes: E2B, E4B, 26B A4B, and 31B. Their diverse sizes make them deployable in environments ranging from high-end phones to laptops and servers, democratizing access to state-of-the-art AI. Gemma 4 introduces key capability and architectural advancements, including reasoning, extended multimodalities, diverse and efficient architectures, optimized for on-device deployment, increased context window, enhanced coding and agentic capabilities, and native system prompt support.

Background and Context

Gemma 4 is built from the same research as Gemini 3 and released under a commercially permissive Apache 2.0 license. This allows developers to use the models for various applications without worrying about licensing restrictions. The release of Gemma 4 marks a significant milestone in the development of open-source AI models, providing a foundation for digital sovereignty and granting teams complete control over their data, infrastructure, and models.

The previous version of Gemma, released in June 2024, was a unified, encoder-free multimodal model designed to bring high-performance multimodal intelligence directly to laptops. The new Gemma 4 models build upon this foundation, offering advanced reasoning capabilities and native audio inputs. The developer community has played a crucial role in the development of Gemma 4, with over 150 million downloads across various applications.

Why it Matters to the Industry

The release of Gemma 4 is significant for several reasons. Firstly, it provides a foundation for digital sovereignty, allowing teams to control their data, infrastructure, and models. This is particularly important in industries where data security and compliance are paramount. Secondly, the advanced reasoning capabilities of Gemma 4 make it an ideal choice for complex tasks such as text generation, coding, and reasoning.

The availability of Gemma 4 on Google Cloud further solidifies its position as a leading platform for AI development. The model can be deployed on Vertex AI endpoints, allowing developers to fine-tune the model using Vertex AI Training Clusters (VTC). This ensures efficient adaptation and deployment of the model across various applications.

What Comes Next

The release of Gemma 4 marks a significant milestone in the development of open-source AI models. As the industry continues to evolve, it is likely that we will see more advanced multimodal intelligence capabilities being developed. The availability of Gemma 4 on Google Cloud and its accessibility for various devices make it an ideal choice for developers looking to integrate AI into their applications.

Key Facts

• Gemma 4 is a family of open models built by Google DeepMind, released under a commercially permissive Apache 2.0 license.
• The new models are multimodal, handling text and image input (with audio supported on small models) and generating text output.
• Gemma 4 features a context window of up to 256K tokens and maintains multilingual support in over 140 languages.
• The models are available in four distinct sizes: E2B, E4B, 26B A4B, and 31B.
• Gemma 4 introduces key capability and architectural advancements, including reasoning, extended multimodalities, diverse and efficient architectures, optimized for on-device deployment, increased context window, enhanced coding and agentic capabilities, and native system prompt support.