Story Content

Google has unveiled its Gemma 4, referring to it as the most intelligent open model of its era. The new family was created based on the lessons of Gemini 3 research and offers frontier-level performance in complex reasoning, agentic behavior, multimodal processing (text, vision, and audio), and high-quality offline code generation.

The highlight is the super-efficient E2B and E4B models, which are explicitly made to operate on smartphones and edge computers such as Raspberry Pi. These miniature specifications can work entirely offline with almost zero latency, complete through optimization with a partner, including Qualcomm and MediaTek. The users are able to do sophisticated work, such as multi-step planning and real-time audiovisual processing, and even at the same time keep all the information confidential to the device.

The bigger 26B Mixture-of-Experts and 31B Dense models serve more demanding usages on both GPUs and servers. Gemma 4 is open-source under the permissive Apache 2.0 license and can be hosted on Hugging Face and Google Cloud to enable developers to create privacy-conscious, low-latency AI applications.

With this release, the company takes a significant step towards on-device intelligence, lessening the need to use cloud services and enabling strong intelligence even when not connected to the internet. Everything with mobile assistants to offline coders can now be tried immediately by system developers and tech enthusiasts on Gemma 4.

AI has been personalized and made more affordable!