Gemma 4 technical report details new open-weight multimodal model

Gemma 4 introduces dense and Mixture-of-Experts architectures for open-weight, natively multimodal language models. The models aim to advance compute efficiency and reasoning.