Google DeepMind releases DiffusionGemma, a model that runs local AI 4x faster

Google has released a new AI model called DiffusionGemma that generates text four times faster than comparable models, making it practical to run on regular gaming computers instead of expensive server hardware. Unlike most AI assistants that write one word at a time, this model thinks through entire blocks of text at once—similar to how image generators work—which is why it's so much quicker. The catch is that it's still quite large, though Google designed it to fit within the memory limits of high-end consumer GPUs.