Google released Gemma, a family of lightweight open-weight models built, it said, from the same research and technology as Gemini. It started with two sizes, 2B and 7B parameters — the 7B aimed at consumer-grade GPUs and TPUs, the 2B at CPUs and on-device use — both licensed for commercial as well as research use. Google was careful to call them "open models," open-weight with terms of use attached rather than open source. It marked Google's entry into the open-weight territory Meta's Llama had held, and settled the pattern common across the industry: keep the frontier model closed, open up the small ones.