Bring any idea to life with multimodal capabilities
Google's largest and most capable AI model. Built from the ground up to be multimodal, Gemini can generalize and seamlessly understand, operate across and combine different types of information, including text, images, audio, video and code.
Gemini 1.0 is optimized in three sizes—Ultra for complex tasks, Pro for scaling across tasks, and Nano for on-device efficiency. This flexibility allows developers and enterprises to deploy the right model size for their infrastructure and performance requirements.
Gemini can understand, explain, and generate code in popular languages like Python, Java, C++, and Go. Its ability to reason about complex information and work across languages makes it effective for developers building and scaling AI applications.
Gemini Ultra achieves 90.0% on MMLU (massive multitask language understanding) and outperforms human experts on 30 of 32 leading academic benchmarks. Its sophisticated reasoning capabilities help solve complex problems in math, physics, law, medicine, and other domains requiring deliberate problem-solving.
Gemini is natively multimodal and pre-trained from the ground up to understand and reason across all modalities seamlessly. Unlike models that stitch separate components together, Gemini's unified architecture enables sophisticated reasoning about complex written and visual information, making it uniquely skilled at extracting insights from diverse data types.
Gemini's multimodal reasoning can read, filter, and understand information across hundreds of thousands of documents. Its ability to uncover knowledge difficult to discern amid vast data enables breakthroughs in fields like science and finance at digital speeds.