Google AI Releases DiffusionGemma, a 26B MoE Open Model Using Text Diffusion for Up to 4x Faster Generation

- Google dropped DiffusionGemma, a 26B parameter model that generates text by painting over noise instead of the tedious left-to-right token-by-token grind. It’s 4x faster on GPUs because it drafts blocks in parallel, not because it’s smarter. In fact, it’s dumber than standard Gemma 4; it solves zero Sudoku puzzles out of the box. It’s a speed demon with a brain of putty. While moonboys chase rugs and regulators chase shadows, Google is just trying to make local inference less painful for devs who hate waiting. It’s a neat trick for rapid iteration, but don’t expect it to predict Bitcoin’s next move or save your wallet from Discord scams. It’s faster, sure, but still fundamentally guessing words while you wait for the market to crash anyway.