Gemini Diffusion: Google DeepMind's Model
Source: blog.google
We are continuously exploring new ways to refine our models, focusing on improved efficiency and performance. Gemini Diffusion, our latest research model, excels as a text diffusion model. It's designed to produce outputs by transforming random noise into understandable text or code, similar to our advanced models for image and video creation.
The Gemini Diffusion experimental demo, launched today, generates content with improved speed relative to our previous models, while maintaining its coding capabilities. If you want to try the demo, please sign up on the waitlist.
We are dedicated to identifying methods to decrease latency across our Gemini models, including a faster 2.5 Flash Lite coming in the future. Stay in touch to receive the latest Google updates.