Google DeepMind·Gemma

Gemma 4 26B MoE

The Gemma 4 MoE activating only 3.8B parameters per step for local speed and intelligence density.

CurrentLanguageReasoningCodingAgentMultimodalOpen weights

MODEL NUMBER106

Specifications

Only published specifications are shown.

Total parameters

26B

Active parameters

3.8B

Architecture

MoE

Context

256K

Input modalities

Text · Image · Video

Output modalities

Text

Open weights

Yes

License

Apache-2.0

Status

Current

The Gemma 4 MoE activating only 3.8B parameters per step for local speed and intelligence density.

PreviousGemma 4 31B31B · 256K ctx

CurrentGemma 4 26B MoE26B · 256K ctx

New capabilities Efficiency

EfficiencyReasoningTool use

2026-04-02Open weights

Gemma 4 26B MoE launched.

Select a card to view the related model.