Gemma 4: Byte for byte, the most capable open modelsGemini / Apr 2, 2026Apache 2.0 open releaseE2B/E4B (edge) and 26B MoE / 31B dense (workstation)128K–256K context, multimodal, function-callinggemmaopen-modelmultimodaledgemoequantizationfine-tuning