DiffusionGemma: 4x faster text generation

DeepMind Blog

DiffusionGemma is a 26B MoE model (3.8B active) that generates text via diffusion, achieving up to 4x speedup over autoregressive Gemma 4 models (1000+ tokens/s on H100, 700+ on RTX 5090). Released under Apache 2.0, it is designed for speed-critical local workflows like in-line editing and code infilling.

GemmaDiffusion

See 14 more sources

NVIDIA Accelerates Google DeepMind’s DiffusionGemma for Local AI11 days agoMichael Fukuyama

Run DiffusionGemma on NVIDIA for Developer-Ready, High-Throughput Text Generation11 days agoAnu Srivastava

DiffusionGemma: The Developer Guide11 days ago

DiffusionGemma is our new experimental open model with up to 4x faster output on dedicated GPUs....11 days agoGoogle DeepMind

RT @googlegemma: Meet DiffusionGemma! An experimental open model that explores a fast approach to...11 days agoHugging Face

Awesome to see this innovation in text diffusion. DiffusionGemma is lightning fast, 4x faster than...9 days agoDemis Hassabis

RT @osanseviero: Introducing DiffusionGemma, our first exploration with open diffusion text...11 days agoHugging Face

DiffusionGemma11 days agoSimon Willison

Google's DiffusionGemma generates 256 tokens in parallel and self-corrects as it goes10 days ago

Google's DiffusionGemma AI Hits 1,000 Tokens Per Second—And It's Free11 days agoJose Antonio Lanz

Google’s DiffusionGemma is 4x faster than its other Gemma models11 days agoFrederic Lardinois

DiffusionGemma: The Developer Guide- Google Developers Blog11 days agotevlonDiscuss

Diffusion Gemma is 4x faster, but makes 6x more mistakes!9 days agogladkosDiscuss

Qwen Who? DiffusionGemma running at 1,500 tk/s on a Digital Pregnancy Test.10 days agoPorespellarDiscuss

DeepMind Blog

Jun 10, 4:24 PM