We present Gemini 1.5
Author: Demis Hassabis, General Director of Google Deepmind, on behalf of the Gemini team
This is an exciting time for artificial intelligence. New progress in this field can make AI more helpful for billions of people in the coming years. From We present Gemini 1.0We tested, improve and increase its capabilities.
Today we announce our new generation model: Gemini 1.5.
Gemini 1.5 provides significantly increased performance. This is a change in our approach, based on innovations in the field of research and engineering in almost every part of our development and infrastructure of the Foundation model. This includes the increase in Gemini 1.5 performance for training and serving with the new A mixture of experts (Maybe) architecture.
The first model Gemini 1.5 that we release for early tests is Gemini 1.5 Pro. It is a medium -sized multimodal model, optimized for scaling in a wide range of tasks and works at a similar level up to 1.0 ultraOur largest model so far. It also introduces a breakthrough experimental function in understanding of the long -term.
Gemini 1.5 Pro has a standard context window of 128,000 tokens. But from today a limited group of programmers and corporate clients can try it with a contextual window up to 1 million tokens through For learning AND Vertex AI In a private preview.
When we introduce a full context window of 1 million tokens, we are actively working on optimizations to improve delays, reduce calculation requirements and improve the user's experience. We are excited that people will try this breakthrough capabilities, and below we provide more details about future availability.
These further progress in our new generation models will open new opportunities for people, programmers and enterprises to create, discover and build with artificial intelligence.
