Models · The Decoder ·
Google's new open model DiffusionGemma generates text from noise instead of word by word
Google released DiffusionGemma, a 26B open model that generates text using diffusion rather than token-by-token decoding. Nvidia says it can reach about 1,000 tokens per second on a single H100 GPU, but Google says quality is lower and is positioning it as an experimental developer tool.