DiffusionGemma
Google released DiffusionGemma, a text diffusion LLM that is 4x faster than autoregressive models. Apache 2.0 license on HuggingFace.
Model: diffusiongemma-26B-A4B-it
Community reports: Decode 104 tok/s with vLLM. GGUF version at 50 tok/s.
URLs: https://pc.watch.impress.co.jp/docs/news/2116352.html | https://huggingface.co/google/diffusiongemma-26B-A4B-it
Source: X(Twitter) @wmoto_ai, @oitamaIoT, @asap2650 (2026-06-11)

