Adapt
Mini toolsDiffusionGemma
By Google
DiffusionGemma is Google’s experimental open text-generation model that applies diffusion-style generation to language instead of generating strictly token by token. It is a 26B total-parameter Mixture-of-Experts model with 3.8B active parameters during inference, built from Gemma 4 and Gemini Diffusion research. The model drafts 256-token blocks in parallel, refines them iteratively, and uses bidirectional attention, making it useful for speed-critical local workflows such as inline editing, rapid iteration, code infilling, mathematical structures, and other non-linear generation tasks. Google says it can generate up to 4x faster than standard autoregressive decoding on dedicated GPUs, but also states it is experimental and lower quality than standard Gemma 4 for production outputs.
New
Multimodal
Gen 3
Overview
DiffusionGemma is Google’s open experimental 26B MoE text-diffusion model for faster local text generation, parallel editing, code infilling, and non-linear text workflows.
About Google
At Google, we think that AI can meaningfully improve people's lives and that the biggest impact will come when everyone can access it.
Industry:
Technology, Information and Internet
Company Size:
190820
Location:
Mountain View, CA, US
Website:
ai.google
Tools using DiffusionGemma
No tools found for this model yet.
Last updated: June 11, 2026
