The Digester
technology

Google launches Gemma 4 for smartphones and data centres

Google released Gemma 4, four open-source models under Apache 2.0 that scale from effective 2 billion parameter edge models to a 31 billion parameter flagship, offering advanced reasoning, vision, audio and long-context support for devices from smartphones to data centre workstations.

Apr 2nd 2026 · United States

Insights

  • Gemma 4 is released under an Apache 2.0 licence and is available in four sizes for developers to use and modify.
  • Models include E2B and E4B for edge and mobile, a 26B MoE mid-range model, and a 31B dense flagship ranked number three on the Arena AI leaderboard.
  • Google says the 31B model outperforms competitors up to 20 times its size on benchmark comparisons.
  • Key capabilities include advanced reasoning, function-calling for agentic workflows, offline code generation on local machines, and native vision and audio processing.
  • Edge models were co-developed with Google Pixel, Qualcomm and MediaTek and can run offline with near-zero latency on phones and single-board devices.
  • Gemma 4 supports long context windows up to 128,000 tokens on edge models and 256,000 on larger models and was trained across more than 140 languages.

Sources