Gemini Diffusion

2025-05-20 · Publisher : Google · Domain : Artificial Intelligence

Google DeepMind unveils Gemini Diffusion, an experimental research model that generates text and code by diffusion, turning noise into text, much faster than its fastest model.

Google DeepMind unveils Gemini Diffusion, an experimental research model that generates text and code by diffusion, turning noise into text, much faster than its fastest model. ### What changes - Diffusion approach applied to text and code, as for image and video. - Significantly faster generation than the fastest Gemini model, with equivalent coding performance according to Google. - Experimental demo available via waitlist. - Google announces in parallel a faster version of Gemini 2.5 Flash-Lite. ### Specifications - Experimental research model, not available in the API at launch. - Access: waitlist for the demo. ### Pricing Pricing not disclosed (research demo). ### Benchmarks No numerical score in the post; the model page on deepmind.google was not used. ### Key takeaways A research direction to watch for reducing generation latency, with no commercial availability in 2025.

Tags : LLM, Google, DeepMind, Recherche, Diffusion