Google releases Gemma 4 open model family in five sizes

Google DeepMind has completed the Gemma 4 lineup: five sizes from E2B for phones to 31B Dense for workstations, all under Apache 2.0. The flagship 26B A4B MoE activates only ~4B parameters per token and needs 14.4 GB at Q4 quantization.

• Five sizes: E2B, E4B, 12B Unified, 26B A4B and 31B Dense
• 26B A4B activates ~4B parameters per token at 14.4 GB in Q4
• 31B Dense: MMLU Pro 85.2, AIME 2026 89.2, LiveCodeBench v6 80.0
• 12B Unified drops the encoder and handles text, images and audio