Google has released an open-source artificial intelligence model with no restrictions on commercial use. The move is interpreted as Google jumping into the open-source AI competition, spurred by the emergence of OpenClaw, an open-source AI agent platform.
Google unveiled Gemma 4, a lightweight open-source AI model, on Friday. The release comes one year after the company introduced Gemma 3 in March last year.
While users must pay a subscription fee to use Gemini, Google's most advanced AI model, Gemma — first released in February 2024 — is distributed free to developers. However, Gemma 4 was built on the same technical foundation as Gemini 3, which launched in November last year.
Gemini is a cloud-based model connected via the internet, whereas Gemma is an on-device model designed to run on personal devices offline. Gemma can be used on supercomputers, laptops and smartphones. Because the large language model (LLM) loaded on the device processes user queries directly without sending them to internet servers, the model offers faster response times and eliminates the risk of data leaks.
The most significant change in Gemma 4 is its adoption of the Apache 2.0 license, which permits commercial use. Until now, U.S. Big Tech companies — with the exception of Meta — have pursued closed-source strategies that withhold core information. Chinese companies such as DeepSeek and Qwen, by contrast, have focused on open-source approaches. As a result, AI startups in Europe and elsewhere have been building enterprise AI agents and chatbot services using LLMs, models and code from DeepSeek and Qwen.
The emergence of OpenClaw earlier this year has drawn particular attention to open-source approaches for AI agents. Nvidia also recently unveiled NemoClaw, its own open-source AI agent platform.
Still, analysts say Google's ultimate intention is to dominate the on-device AI agent market by integrating Gemma with Android, its mobile operating system. Google emphasized that "Gemma 4 can run on billions of Android devices and select laptop GPUs." Nvidia's NemoClaw is similarly open-source with publicly available source code, but Nemotron — the model required to use NemoClaw — is designed to perform most efficiently when combined with Nvidia's chips and software.