Gemma 4 is Google's open model family released in April 2026 under the Apache 2.0 licence, with sizes from phones up to workstations.
In plain words
Gemma 4 is built from the same research as Gemini 3. It launched on 2 April 2026 in four sizes, from small 'effective 2B' and '4B' models for phones and browsers to a 31B dense model and a 26B mixture-of-experts model; Google's docs now also list a 12B model. All can reason with an adjustable thinking mode, read images and video, and the smaller ones also take audio. Small models handle 128K tokens of context and the medium ones 256K.
A quick example
Imagine a developer building an offline translation app for Android. They could pick the E2B model, which needs roughly 1 GB of memory on a phone, and keep a larger Gemma 4 model on a workstation for harder jobs.
Where you'll meet it
- On-device and in-browser AI.
- Local agents with function calling.
- Fine-tuning for specialised tasks.
Related
Sources
- DocsGemma 4: our most intelligent open models, Google · read 29 Sept 2026
- DocsGemma 4 model overview, Google AI for Developers · read 29 Sept 2026
- DocsGemma 4 model card, Google AI for Developers · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.