Gemma 4 expands Google’s open-model family

Google DeepMind introduced Gemma 4, a family of open-weight models released under Apache 2.0. The lineup emphasizes reasoning, multimodality, and agentic workflows, offering an option for teams that need their own deployment alongside Google’s hosted Gemini products.

Context

A local-model choice begins with the device and the job. Smaller variants can suit frequent requests, while larger ones may help with difficult cases. Downloadable weights provide control over placement, but maintenance and quality evaluation still belong in the operating budget.