Prime Highlights
- Google DeepMind launches Gemma 4, a family of four open-weight AI models built to run on devices from smartphones to workstations.
- The 31B Dense model ranks third among open models on the Arena AI leaderboard despite its compact size.
Key Facts
- All Gemma 4 models release under an Apache 2.0 license, allowing free use and modification.
- Developers download the Gemma series over 400 million times since its launch.
Background
Google DeepMind launches Gemma 4, a new family of four open-weight artificial intelligence models designed to run on devices ranging from small edge endpoints to full workstations, making advanced AI more accessible to developers worldwide.
The four models in the lineup are the Effective 2B, Effective 4B, a 26B Mixture of Experts, and a 31B Dense model. All four come under an Apache 2.0 license, which removes earlier restrictions and gives developers far greater freedom to use, modify, and redistribute the models.
The 31B Dense model currently sits third among open models on the Arena AI text leaderboard, while the 26B Mixture of Experts holds sixth place. Google says both models outperform others up to twenty times their size in parameter count, a notable achievement given their relatively compact build.
The smaller Effective 2B and 4B models target edge devices such as smartphones, Raspberry Pi, and Nvidia Jetson hardware. The models operate through complete offline functionality, which delivers immediate performance while processing native audio input. The larger 26B and 31B models support context windows of up to 256,000 tokens and fit on a single 80GB NVIDIA H100 GPU.
Since Google first introduced Gemma to the open-source community in February 2024, developers have downloaded the series more than 400 million times and created over 100,000 community variants, which shows that there is strong global interest in the series.
The Apache 2.0 License also expands the development ecosystem by providing day-one support for many popular tools such as Ollama, vLLM, and LM Studio.
Gemma 4 is presently accessible through Google AI Studio, Kaggle, Hugging Face and Ollama, while its production deployment works with Google Cloud’s Vertex AI and Cloud Run.