Google Launches Open Gemma 4 Models with Multilingual Support and Apache 2.0 License

• April 3, 2026

Google has unveiled a new family of open-source AI models named Gemma 4, expanding its artificial intelligence offerings with a broad focus on multilingual capabilities and device compatibility. These models support 140 languages and are distributed under the Apache 2.0 license, allowing for flexible use across various applications.

Versatile Models Tailored for Diverse Hardware

Developed as an evolution of the company’s earlier Gemini 3 technology introduced late last year, the Gemma 4 series features four distinct models differentiated primarily by parameter size. This diversity aims to address the performance requirements of different hardware environments, ranging from resource-constrained devices to high-performance computational systems.

One subset, labeled as Effective, targets devices with limited processing power, including smartphones. These models come in smaller scales of 2 billion and 4 billion parameters, allowing for more efficient execution without heavy resource demands. In contrast, larger-scale Gemma 4 models accommodate more powerful platforms: a 26 billion parameter version utilizing a Mixture of Experts (MoE) architecture, and a 31 billion parameter Dense model designed for comprehensive performance.

The Mixture of Experts approach incorporated into one of the Gemma 4 variants is known for selectively activating parts of the network during inference, which can improve efficiency at scale. Meanwhile, the Dense configuration involves fully activated layers during processing, often resulting in more consistent output at the cost of increased computational load.

Google’s provision of these models under the Apache 2.0 license suggests an intent to promote widespread adoption and integration within the developer and research communities. The broad language support further enhances the models’ applicability, making them suitable for global applications in natural language processing, translation, and conversational AI.

With the Gemma 4 family, Google continues to emphasize transparency and accessibility in AI, supporting a range of devices from personal mobile hardware to enterprise systems. Details on deployment scenarios, performance benchmarks, or pricing were not disclosed.

The introduction of these models represents a noteworthy advancement in open AI frameworks, signaling ongoing efforts from major technology companies to balance computational demands with versatile multilingual functionality.

Google introduces Gemma 4, a family of open AI models supporting 140 languages and designed for diverse device capabilities.