Google's experimental DiffusionGemma (26B parameters) generates 256 tokens simultaneously instead of sequentially, achieving 4x faster output, over 1,000 tokens/sec on an H100, though with lower quality than standard Gemma 4.

Get hand-picked daily summaries of the best, most informative AI articles from around the web.

Google released DiffusionGemma, a 26-billion parameter experimental AI model that generates entire blocks of 256 tokens at once instead of word by word, delivering up to 4x faster text output, exceeding 1,000 tokens per second on an NVIDIA H100. It's designed for speed-focused local workflows, though output quality is lower than standard Gemma 4.

DiffusionGemma: 4x faster text generation

Get the next batch of curated summaries in your inbox.