Too intense, Google dropped a bombshell early this morning. They unveiled their latest open-source model, Gemma 4, emphasizing that it can run locally on personal devices. Unlike relying on cloud-based calls, Gemma 4 supports running on users' local hardware while offering function-calling capabilities. This means the model can not only chat but also directly use tools, browse the web, execute code, and even connect to APIs—essentially functioning as an AI agent. The key shift here is that inference costs have been drastically reduced, almost to zero. Users no longer need to pay for each call, only bearing the cost of their local hardware. Market analysts believe this will disrupt the pay-per-call AI business model, especially for companies like OpenAI and Anthropic that rely on API-based revenue. The related Chinese documentation page has also been released simultaneously.