FrontierThe story, in brief
Google's new open model DiffusionGemma generates text from noise instead of word by word
4x faster. Google's DiffusionGemma trades quality for speed—a fundamental shift in how text generation works.
The KeyNews take
Why it matters
Google is experimenting with diffusion-based text generation as an alternative to autoregressive models, achieving 4x speed gains on H100s but with quality tradeoffs. This signals exploration of fundamentally different generation architectures that could reshape inference economics.
The key facts
6 to knowDiffusionGemma: 26B parameters
Diffusion-based generation (noise-to-text, not token-by-token)
~1,000 tokens/second on single H100 GPU
4x faster than comparable autoregressive models
Lower output quality vs. standard models
Positioned as experimental/developer tool
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: Google released DiffusionGemma, a 26-billion-parameter model that generates text not token by token but through diffusion, similar to how image AI turns noise into a picture. According to Nvidia, it hits about 1,000 tokens per second on a single H100 GPU, roughly four times faster than comparable…