Google DeepMind·Gemma

DiffusionGemma 26B

An experimental Gemma 4 text-diffusion model exploring lower latency through parallel block generation.

CurrentLanguageOpen weights

MODEL NUMBER108

Specifications

Only published specifications are shown.

Total parameters

26B

Architecture

MoE

Input modalities

Text

Output modalities

Text

Open weights

Yes

License

Apache-2.0

Status

Current

An experimental Gemma 4 text-diffusion model exploring lower latency through parallel block generation.

PreviousGemma 4 26B MoE26B · 256K ctx

CurrentDiffusionGemma 26B26B · — ctx

New capabilities Text diffusion Fast action Open research

Text diffusionFast actionOpen research

2026-06-10Open weights

Google released experimental DiffusionGemma.

Select a card to view the related model.