DiffusionGemma: 4x Faster Text Generation

Google has introduced DiffusionGemma, an experimental 26B Mixture of Experts model that uses text diffusion to generate text up to 4x faster than traditional autoregressive models by producing entire blocks of text simultaneously rather than token-by-token. The model achieves 1000+ tokens per second on NVIDIA H100 GPUs and uses bidirectional attention to enable parallel generation, making it suitable for speed-critical applications like real-time editing and code infilling, though it sacrifices some output quality compared to standard Gemma 4 models.