Developers talk about high performance in a variety of tasks.
On March 22, Google opened access to its powerful Gemini 1.5 Pro neural network, which allows processing up to a million tokens. Jeff Dean, the company's leading researcher, announced the gradual connection of users to the API, while developers can already test the neural network through AI Studio.
The innovative model has high performance and is used in various tasks, including historical data analysis and code processing.
For experienced use of Gemini 1.5 Pro you need:
Open Google Chrome and activate VPN, with the region Singapore or USA. Go to aistudio.google.com/app. Login via Gmail.
For example, when analyzing “War and Peace,” the model completed the task in just 305 seconds. Some users note the need to link a Google account to an authorized country, although there are reports of access to the model without this procedure. You can also create a new account using a virtual SIM card.
Gemini 1.5 Pro is focused on processing large amounts of text information, work data, video and audio. The Google-tested model achieved a 99% success rate when tested in the NIAH (Needle In A Haystack) test and demonstrated adaptive learning capabilities in the MTOB (Machine Translation from One Book) benchmark.
The promise of using the technology is enormous: thanks to the specific Transformer and MoE architecture, Gemini 1.5 Pro is capable of not only efficiently handling data sets, but also demonstrates adaptive learning capabilities comparable to the previous Gemini 1.0 Ultra model. Google continues to actively research and expand the functionality of its AI platform.