Google has announced that it’s releasing Gemini 1.5 Pro for early testing via AI Studio. One of the key focuses with this new model update is its ability to understand long-context content supporting a 128,000 token context window at first but ambitions to scale this up to 1 million tokens.
According to a technical paper published by Google, Gemini 1.5 Pro performs better than both 1.0 Pro and 1.0 Ultra. Compared to Ultra, 1.5 Pro has a Core Capabilities win-rate of 54.8%, a Text win-rate of 77%, a Vision win-rate of 46%, and an Audio win-rate of 20%. It performs much better than 1.0 Pro scoring 87.1%, 100%, 77%, and 60%, respectively.
Some of the highlights of this model include:
Reasoning about vast amounts of information
Gemini 1.5 Pro can analyze and summarize the 402-page transcripts from Apollo 11’s mission to the moon.
Better understanding across modalities
Gemini 1.5 Pro can perform highly sophisticated reasoning tasks for different modalities, like a silent Buster Keaton movie.
Problem-solving with longer blocks of code
Gemini 1.5 Pro can reason across 100,000 lines of code giving helpful solutions, modifications and explanations.
Google CEO Sundar Pichai had this to say about the new version of Gemini:
"Our teams continue pushing the frontiers of our latest models with safety at the core. They are making rapid progress. In fact, we’re ready to introduce the next generation: Gemini 1.5. It shows dramatic improvements across a number of dimensions and 1.5 Pro achieves comparable quality to 1.0 Ultra, while using less compute.
This new generation also delivers a breakthrough in long-context understanding. We’ve been able to significantly increase the amount of information our models can process — running up to 1 million tokens consistently, achieving the longest context window of any large-scale foundation model yet."
If you would like to gain early access to Gemini Pro 1.5, fill in this Google AI Studio form. Google didn’t give a timeline for when it would release this update to all users, but it’ll be whenever it has completed its testing.
Source: Google