Gemma 4 open models are now available, and Google says this is its most intelligent open model family to date. Built on the same research foundation as Gemini 3, the new Gemma 4 family is designed for advanced reasoning, agentic workflows, and on-device performance across a wide range of hardware. Since the first Gemma launch, developers have downloaded the models over 400 million times, creating more than 100,000 variants in what Google calls the “Gemmaverse.”

Four Sizes, One Family: What Gemma 4 Offers

Google is releasing Gemma 4 in four configurations: Effective 2B (E2B), Effective 4B (E4B), a 26B Mixture of Experts (MoE), and a 31B Dense model. Each is built for a different hardware tier. The 31B model currently ranks third among all open models on the Arena AI text leaderboard, while the 26B MoE holds the sixth spot. Both outperform models up to 20 times their size. For developers working in artificial intelligence, that kind of efficiency matters enormously.

The E2B and E4B edge models are built specifically for mobile and IoT devices. They activate only 2 billion and 4 billion parameters during inference, preserving RAM and battery life. Developed in collaboration with Google Pixel, Qualcomm Technologies, and MediaTek, these models run fully offline with near-zero latency on phones, Raspberry Pi, and NVIDIA Jetson Orin Nano devices.

Key Capabilities Across the Gemma 4 Open Models

  • Advanced reasoning: Multi-step planning and deep logic, with strong results on math and instruction-following benchmarks.
  • Agentic workflows: Native function-calling, structured JSON output, and system instructions for building autonomous agents.
  • Code generation: High-quality offline code support, turning any workstation into a local AI coding assistant.
  • Vision and audio: All models process video and images natively. E2B and E4B also support audio input for speech recognition.
  • Long context: Edge models support 128K context windows; larger models go up to 256K.
  • 140+ languages: Natively trained across more than 140 languages for global application development.

Apache 2.0 License: A Major Shift for Developers

One of the most significant announcements is the licensing. Gemma 4 is released under an Apache 2.0 license, giving developers full commercial freedom with no restrictive barriers. This means complete control over data, infrastructure, and model deployment, whether on-premises or in the cloud. For enterprises and sovereign organizations in the technology sector, this is a meaningful step toward digital independence.

“The release of Gemma 4 under an Apache 2.0 license is a huge milestone. We are incredibly excited to support the Gemma 4 family on Hugging Face on day one.”

Hugging Face

Real-world applications are already emerging. INSAIT used earlier Gemma models to build BgGPT, a Bulgarian-first language model. Yale University collaborated with Google on Cell2Sentence-Scale to explore new pathways for cancer therapy. These examples show what becomes possible when capable AI models are made openly available.

Where to Access and Deploy Gemma 4

Developers can start experimenting immediately. The 31B and 26B MoE models are available in Google AI Studio, while the E2B and E4B models are accessible through Google AI Edge Gallery. Android developers can use Gemma 4 to power Agent Mode in Android Studio, with production support via the ML Kit GenAI Prompt API.

The models are available for download from Hugging Face, Kaggle, and Ollama. Supported tools include Transformers, vLLM, llama.cpp, MLX, LM Studio, Keras, and many others. For teams building applications at scale, Google Cloud offers deployment through Vertex AI, Cloud Run, GKE, and TPU-accelerated serving. Google has also launched the Gemma 4 Good Challenge on Kaggle, inviting developers to build products with meaningful real-world impact.