Google Gemma 4 release: 26 billion parameter open source model, local equipment can run

🚀 Google's latest open source · local running · completely free

Google just released it. Gemma 4 A series of open source models, the latest multi-modular, multi-linguistic model family of Google DeepMind. The most attractive part is Gemma 4 not onlyFull open source free.It's still there.Local device offlineThat means you can have a powerful AI assistant on your computer, even your cell phone, without the Internet, completely private data.

---

📊 Gemma 4 Series Model at a glance

Gemma 4 Models with four different specifications covering full deployment scenarios from mobile phones to data centres

Model	Structure	Total parameters	Valid parameters	Context	Model
Gemma-4-E2B	Dense.	5.1B	2.3B	128K	Text + Image + Audio
Gemma-4-E4B	Dense.	7.9B	4.5B	128K	Text + Image + Audio
Gemma-4-26B-A4B	MoE.	26B	3.8B	256K	Text + Image
Gemma-4-31B	Dense.	31B	31B	256K	Text + Image

Gemma 4 Four Model Specifications

of which E2B and E4B It is a lightweight model specially designed for peripheral equipment, which can operate on low-capacity equipment such as mobile phones, strawberry pies, etc.;26B-A4B Using MoE (mixed experts) architecture, activation parameters are only 3.8B, but performance is close to 31B;31B This is the strongest Gemma 4 model.

---

Four core spots.