Leading Google researcher Jeff Dean announced the opening of public access to the Gemini 1.5 Pro neural network. The proprietary language model copes with a wide range of tasks and, according to its creators, significantly exceeds GPT-4 in its capabilities.
Gemini 1.5 Pro was the first to cross the million token threshold in a single request (approximately 700,000 words). The neural network is capable of processing up to 11 hours of audio and up to 1 hour of video within one request, reading events that occur in the text and in the frame. AI can analyze up to 30,000 lines of code and understands what is shown in photographs, determining the context and objects.
According to Google, the Gemini 1.5 Pro model not only outperformed GPT-4 by 8.2 times in the size of the “context window”, but was also able to understand information throughout its entire length.
Red marks – Gemini 1.5 Pro errors, gray marks – GPT-4 lag behind Gemini 1.5 Pro
Access to the neural network is currently only available in some regions and in a limited number of languages. You can view the “system requirements” of Gemini 1.5 Pro at this link .