The Google Research team introduced the experimental DiffusionGemma model based on discrete diffusion
DiffusionGemma, a model created by fine-tuning Gemma 4, generates text through discrete diffusion in parallel blocks of 256 tokens and, according to the authors, achieves about 1 500 tokens per second on a single NVIDIA H100 GPU.
Companies running LLM inference at high volumes may infer from the results that diffusion decoding could reduce compute costs and generation latency if this approach becomes more widely adopted in production deployments.