Key takeaways
- Google rolls out API support for Gemini 1.5 Pro, enhancing developer access to advanced AI capabilities.
- Gemini 1.5 Pro processes up to 10 million tokens, revolutionizing information handling in various formats.
- The model combines Transformer and MoE architecture, optimizing performance across diverse tasks.
- Achieves a 99% success rate in locating facts within lengthy texts, showcasing its precision.
- Follows the release of Gemini Ultra and integrates new generative AI features into Google Chrome.
Jeff Dean, chief scientist at Google DeepMind, took to X to share that the company is starting to roll out API support for Gemini 1.5 Pro.
Gemini 1.5 Pro comes with a standard 128,000 token context window. It can process vast amounts of information in one go, including one hour of video, 11 hours of audio, codebases with over 30,000 lines of code, or over 700,000 words. In their research, Google also successfully tested up to 10 million tokens.
“We'll be onboarding people to the API slowly at first, then ramping it up. In the meantime, developers can try out Gemini 1.5 Pro in the AI Studio UI right now,” said Dean.
Gemini 1.5, using Transformer and MoE architecture, combines the strengths of both models. Traditional Transformers function as one large network, whereas MoE divides models into smaller "expert" networks. Gemini 1.5 Pro excels in various tasks, such as analysing historical transcripts like Apollo 11's mission and understanding silent movies. It efficiently processes extensive code, showcasing adaptability.
Notably, the Needle In A Haystack (NIAH) evaluation achieves a 99% success rate in locating specific facts within long texts. Its ability to learn in context, shown in the Machine Translation from One Book (MTOB) benchmark, establishes it as a leader in adaptive learning.
This development follows Google's release of Gemini Ultra. Additionally, Google integrated generative AI features into Chrome and introduced the "Help Me Write" feature across all websites.
