DiffusionGemma

Simon Willison · Jun 10, 2026 · Product Launches

Google released DiffusionGemma, an open-weight Apache 2 diffusion-based text generation model.

“That research has returned in the best possible way: as a new open weight (Apache 2 licensed) Gemma model”

Google turned its experimental Gemini Diffusion research into DiffusionGemma, an Apache 2 open-weight diffusion text model now hosted free on NVIDIA's NIM API. Diffusion architectures matter because they generate at 500+ tokens/second, far faster than autoregressive models, and an open-weight release lets the community build on a fundamentally different generation approach.

google open-weights diffusion-llm gemma inference-speed

Watch / read the original source →