This is Gemma 4, the most advanced and intelligent version of Google’s open AI for developers.

Google has introduced Gemma 4, the most powerful and latest addition to the Gemma family, its open AI models designed for developers and researchers. In developing this updated version, the technology company has relied on the same advanced technology and cutting-edge research that it utilized for Gemini 3, granting it a superior level of intelligence.

Gemma 4 has been engineered specifically to address advanced reasoning tasks and agent-based workflows. Additionally, this AI has been released in four different sizes, and is available under the Apache 2.0 license, which allows for commercial use.

As explained by Google: “This open-source license provides a foundation for complete developer flexibility and digital sovereignty, granting you full control over your data, infrastructure, and models. It enables you to develop freely and deploy securely in any environment, whether locally or in the cloud.”

Gemma 4 delivers open, cutting-edge performance and intelligence

Gemma 4 operates efficiently and is optimized for all types of hardware, ranging from Android devices to laptop GPUs, workstations, or accelerators for developers.

“By utilizing these highly optimized models, you can fine-tune Gemma 4 to achieve state-of-the-art performance on your specific tasks. We have already witnessed remarkable success with this approach; for instance, INSAIT created a pioneering language model that prioritizes Bulgarian (BgGPT), and we collaborated with Yale University on Cell2Sentence-Scale to discover new approaches for cancer therapy, among many other projects,” states Google.

Likewise, the capabilities that make Gemma 4 the most advanced branch of this family of models are as follows:

• It has been natively trained in over 140 languages: developers can leverage Gemma 4 to create high-performance applications for a global audience.
• Advanced reasoning: this AI can plan across multiple steps and develop deep logic, showing substantial improvements in mathematical tests and instruction-following.
• Agent-based workflows: native support for function calling, structured JSON output, and native system instructions enable the creation of autonomous agents capable of interacting with various tools and APIs and reliably executing workflows.
• Supports high-quality offline code: this transforms your workstation into an AI code assistant that prioritizes local processing.
• Vision and audio: all of its models natively process video and images, support variable resolutions, and excel at visual tasks such as optical character recognition (OCR) and graphical comprehension. It is important to note that the E2B and E4B models include native audio input for speech recognition and comprehension.
• Broader context: Edge models have a context window of 128 KB, while larger models offer up to 256 KB, enabling you to send extensive content in a single request.

Four versatile sizes

To ensure that each developer can use the model best suited to their hardware and performance requirements, Google has released four versions of Gemma 4 in different sizes. All go beyond simple chat capabilities to handle complex logic and agent-based workflows.

• Dense 31B: The largest model, it delivers cutting-edge performance even without an internet connection. It currently ranks third on the text leaderboard at Arena AI.
• Mixture of Experts 26B (MoE): This model currently ranks sixth on the Arena AI text leaderboard and also provides superior offline performance.
• Efficient 4B (E4B): Focuses on multimodal capabilities, low-latency processing, and system integration over sheer parameter count.
• Efficient 2B (E2B): A more streamlined version of E4B, yet highly efficient.

High security standards

Google guarantees that all models comprising Gemma 4 have undergone the same infrastructure security protocols as its other AI systems, such as Gemini.

“By choosing Gemma 4, businesses and government organizations obtain a secure and transparent foundation that provides advanced capabilities, meeting the highest standards of security and reliability,” the technology company asserts.

Photo: Google