Google Releases Apache 2.0 DiffusionGemma 26B-A4B for Faster Text Generation

Released June 10, 2026, DiffusionGemma 26B-A4B uses text diffusion to generate 256 tokens in parallel. The 26B MoE activates 3.8B parameters and reportedly exceeds 1,000 tokens per second on a single H100, making it up to four times faster than standard Gemma 4. It trails standard Gemma 4 on benchmarks such as MMLU and coding. The model is available on Hugging Face, Kaggle, and Vertex AI Model Garden.

Visit original source