DiffusionGemma

2026-06-10 · Source: Simon Willison's Weblog · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Intermediate, quick

Summary

Google has released DiffusionGemma, a new open-weight model under an Apache 2 license, available as "google/diffusiongemma-26B-A4B-it" on Hugging Face. This release follows an experimental Gemini Diffusion model briefly previewed in May 2025, which achieved 857 tokens/second. The current DiffusionGemma research, now openly available, demonstrates strong performance, with a test generating 2,409 tokens in 4.4 seconds, equating to at least 500 tokens/second. NVIDIA is currently providing free hosting for this model via its NIM cloud API, making it accessible for developers. This marks a significant return for the previously unannounced Gemini Diffusion research, now in an open and usable format.

Key takeaway

For AI Engineers evaluating new text generation models, DiffusionGemma offers a compelling, openly licensed option. You can integrate "google/diffusiongemma-26B-A4B-it" into your projects, leveraging its demonstrated high token generation rates. Consider utilizing NVIDIA's free NIM cloud API for initial testing and deployment, which provides an accessible way to experiment with this model without immediate infrastructure costs. This release enables you to explore advanced text generation capabilities with an open-source foundation.

Key insights

Google's experimental Gemini Diffusion model is now an open-weight, Apache 2 licensed DiffusionGemma.

Principles

• Open-weight releases foster wider adoption.
• Free API hosting lowers access barriers.
• Iterative research can lead to public models.

In practice

• Access DiffusionGemma via Hugging Face.
• Utilize NVIDIA NIM for free inference.
• Benchmark open-weight model performance.

Topics

• DiffusionGemma
• Open-weight Models
• Text Generation
• Apache 2 License
• NVIDIA NIM
• Hugging Face

Best for: NLP Engineer, AI Scientist, Research Scientist, AI Engineer, Machine Learning Engineer, Software Engineer

Related on AIssential

• Run DiffusionGemma on NVIDIA for Developer-Ready, High-Throughput Text Generation — DiffusionGemma employs diffusion-based denoising to generate text tokens in parallel, achieving high throughput for real-time AI.
• Google's DiffusionGemma generates 256 tokens in parallel and self-corrects as it goes — DiffusionGemma applies image diffusion to text, enabling parallel, self-correcting block generation for faster inference in specific contexts.
• NVIDIA Accelerates Google DeepMind’s DiffusionGemma for Local AI — DiffusionGemma uses parallel diffusion for text generation, achieving 4x faster local inference than autoregressive LLMs.
• Google's new open model DiffusionGemma generates text from noise instead of word by word — DiffusionGemma generates text in parallel blocks from noise, significantly boosting local inference speed for non-linear tasks.
• A Visual Guide to DiffusionGemma — DiffusionGemma uses iterative diffusion to generate text, optimizing single-user latency by processing 256 tokens concurrently.
• Generating text with diffusion (and ROI with LLMs) — Diffusion language models offer faster, more accurate text generation via multi-token processing.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Simon Willison's Weblog.