An experiment for researchers chasing fast local text generation; Google admits output quality trails standard Gemma 4.
Kenning's take
What it does
DiffusionGemma is an experimental open model from Google that generates text in blocks using diffusion rather than one token at a time. Google says it is up to four times faster on dedicated GPUs, though output quality is lower than standard Gemma 4. It is a 26B mixture-of-experts model with 3.8B active parameters, released under Apache 2.0 on Hugging Face.
- Generates 256 tokens in parallel per pass
- Up to 4x faster on GPUs, per Google
- 26B MoE with 3.8B active parameters
- Fits in about 18GB of VRAM when quantised
- Apache 2.0 licence
Who it suits. Researchers and developers testing fast local text generation for editing, infilling and interactive tools.
Key facts
- Category
- AI models
- Made by
- Released
- 10 June 2026
- How to get it
- open weights on Hugging Face
- Works on
- open weights
- In the UK
- Yes
- Free plan
- open weights, Apache 2.0
FAQ
When was DiffusionGemma released?
Google announced DiffusionGemma on 10 June 2026.
Is DiffusionGemma better than Gemma 4?
It is faster but Google says its output quality is lower than standard Gemma 4, which it still recommends for top quality.
Is DiffusionGemma free?
Yes. The weights are on Hugging Face under the Apache 2.0 licence.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools
ACE-Step 1.5
Open-source music generation model that writes full songs with lyrics and runs on ordinary consumer GPUs.
AlphaGenome Atlas
Google DeepMind database of AI-predicted effects for every possible single-letter change in human DNA.
Amazon Nova 2
Amazon's second family of Nova AI models, covering reasoning, speech conversation and combined text-and-image generation.
Cisco Antares
Cisco's small open-weight models that find where known security vulnerabilities sit in a codebase, running locally.