Google unveils DiffusionGemma, an AI model that breaks free of left-to-right processing

▤ Summary

Google has launched DiffusionGemma, an experimental open-source AI model that uses diffusion to generate text in parallel instead of token by token. The model claims up to 4x faster inference and is designed to improve local, low-latency workflows on GPUs and TPUs. It targets use cases such as interactive coding, editing, multimodal tasks, and real-time customer support, while also supporting cloud and on-prem deployment options. Google also highlights trade-offs: the model fits best in low-to-medium batch scenarios and delivers lower output quality than standard Gemma 4 in quality-sensitive apps.

▦ Linked Companies

NVIDIA Corporation

$1B+

Hugging Face, Inc.

$10M to $25M