Google's DiffusionGemma Reaches 1,000 Tokens Per Second
Google's DiffusionGemma is a free open AI model that reportedly generates text at up to 1,000 tokens per second by avoiding word-by-word output. The tradeoff is practical: the model still does not run on most people's machines.
What happened?
Google's DiffusionGemma is a free open AI model that reportedly generates text at up to 1,000 tokens per second by avoiding word-by-word output. The tradeoff is practical: the model still does not run on most people's machines.
Why it matters
The development matters because speed is one of the main limits on how AI tools feel in real use. Faster text generation can make AI systems more responsive for companies, developers, and users building products around automated writing, coding, research, and support workflows.
Google has introduced DiffusionGemma, a free open AI model that can generate text at up to 1,000 tokens per second. The model reaches that speed by taking a different approach from typical text generators, avoiding the usual word-by-word generation process.
The development matters because speed is one of the main limits on how AI tools feel in real use. Faster text generation can make AI systems more responsive for companies, developers, and users building products around automated writing, coding, research, and support workflows.
DiffusionGemma's core distinction is its generation method. Instead of producing text sequentially, it uses a diffusion-based approach, which is presented as the reason it can hit the 1,000-token-per-second mark.
That performance does not yet mean the model is accessible to everyone in practice. According to the source material, DiffusionGemma still does not run on most people's machines, limiting how broadly users can test or deploy it today.
For now, DiffusionGemma stands out as a high-speed, free open AI release from Google, but its impact will depend on whether the hardware barrier becomes less restrictive over time.