Google Launches Gemma 4: A New Family of Open-Source AI Models

Original: Google releases Gemma 4, a family of open models built off of Gemini 3

Engadget•Apr 2, 20262 min

In a Nutshell

Google has unveiled Gemma 4, a new family of open-weight AI models built on the technology of Gemini 3. The models, available in four versions, cater to various hardware capabilities and are designed for tasks like image processing and offline code generation, all under an open-source license for greater flexibility.

Key Points

01Gemma 4 includes four versions with varying parameters: 2 billion, 4 billion, 26 billion, and 31 billion.
02The models excel in processing video, images, and audio inputs, making them versatile for various applications.
03Gemma 4 is trained in over 140 languages, enhancing its usability across different regions.
04The open-source release under the Apache 2.0 license allows developers to modify and deploy the models freely.
05Google claims Gemma 4's models outperform larger competitors on AI text leaderboards.

In-Article Ad

Full Summary

Google has introduced the Gemma 4 family of open-weight AI models, building on the advancements of its Gemini 3 Pro. This new lineup features four versions, including the 2-billion and 4-billion parameter models for edge devices, and the more powerful 26-billion and 31-billion parameter models for advanced machines. These models are designed to perform various tasks, including video and image processing, optical character recognition, and offline code generation. The 31-billion and 26-billion variants have earned recognition on Arena AI's text leaderboard, outperforming models that are 20 times their size. Trained in over 140 languages, Gemma 4 aims to provide developers with the flexibility to modify and deploy the models securely, thanks to its release under the Apache 2.0 license. Users can access the model weights through platforms like Hugging Face, Kaggle, and Ollama.

Key Figures

4

Different versions of Gemma 4 released

140

Languages Gemma 4 is trained in

20

Times larger models that Gemma 4 outperformed

In-Article Ad

Why It Matters

The open-source nature of Gemma 4 allows developers and businesses to customize AI solutions without the constraints of proprietary software, potentially leading to innovative applications in various fields.

Who is affected

Developers, businesses, and researchers in AI and machine learning.

What to know

Developers interested in utilizing Gemma 4 should explore its capabilities and consider integrating it into their projects.

FAQ

Gemma 4 includes four versions: 2 billion, 4 billion, 26 billion, and 31 billion parameters.

The model weights for Gemma 4 are available through platforms like Hugging Face, Kaggle, and Ollama.

✦