Google Releases DiffusionGemma: 4x Faster Text Generation with Text Diffusion

Originally published on Google AI Blog — June 10, 2026

Summary

Google introduced DiffusionGemma, an experimental 26B Mixture of Experts open model that uses text diffusion to generate entire blocks of text simultaneously instead of token-by-token, delivering up to 4x faster inference on GPUs. Released under Apache 2.0, the model activates only 3.8B parameters during inference and fits within 18GB VRAM on consumer GPUs, reaching 1000+ tokens per second on an NVIDIA H100.

Key Details

• Architecture: 26B total MoE model, only 3.8B active parameters per inference step.
• Speed: Up to 4x faster than autoregressive models; 1000+ tok/s on H100, 700+ on RTX 5090.
• Memory: Fits within 18GB VRAM when quantized — runs on high-end consumer GPUs.
• License: Apache 2.0 — fully open and available for research and development.
• Use Cases: In-line editing, rapid iteration, non-linear text structures, speed-critical local workflows.
• Built on Gemma 4 family intelligence and Gemini Diffusion research.

Read Original

→ Read full article on Google AI Blog

Curated by Brain Bot for Abhay’s KB — June 12, 2026