Google Launches Gemma 4: Most Capable Open AI Models to Date

Google has launched Gemma 4, the fourth generation of its open AI models, delivering breakthrough intelligence-per-parameter and native multimodal capabilities.

Record-Breaking Performance

Gemma 4 comes in four sizes: Effective 2B (E2B), Effective 4B (E4B), 26B Mixture of Experts (MoE), and 31B Dense. The largest 31B model ranks as the world's #3 open model on the Arena AI leaderboard, while the 26B model takes the #6 spot - outperforming models 20 times their size.

"Built from the same world-class research and technology as Gemini 3, Gemma 4 is the most capable model family you can run on your hardware," Google announces.

Apache 2.0 and Commercial Freedom

One of the biggest changes from previous Gemma generations is that Gemma 4 is released under the Apache 2.0 license - fully open source and commercially permissive. This provides developers with complete flexibility and digital sovereignty.

Multimodal Capabilities

All Gemma 4 models natively process video, images, and audio. Edge models (E2B and E4B) include built-in audio input for speech recognition, while all models support:

• Advanced reasoning: Multi-step planning and deep logic
• Agentic workflows: Native function calling and JSON output
• Code generation: High-quality offline code assistance
• 140+ languages: Natively trained on over 140 languages
• Long context: 128K tokens for edge models, 256K for larger models

Edge AI Revolution

E2B and E4B models are specifically designed for edge devices and run completely offline on phones, Raspberry Pi, and NVIDIA Jetson Orin Nano with near-zero latency.

In close collaboration with the Google Pixel team and mobile hardware leaders like Qualcomm and MediaTek, these models enable agentic workflows directly on device.

Broad Platform Support

Gemma 4 is supported from day one by:

• Hugging Face (Transformers, TRL)
• NVIDIA (NIM, NeMo)
• Ollama, LM Studio, MLX
• Google Cloud (Vertex AI, Cloud Run, GKE)
• Android Studio with Agent Mode

Industry Significance

With over 400 million downloads of previous Gemma generations and an active "Gemmaverse" of over 100,000 variants, Gemma 4 represents Google's ambition to democratize advanced AI.

The commercially friendly Apache 2.0 license and strong performance could challenge both closed models and other open alternatives in the enterprise AI market.