A few days ago, Google's official document said that gemini 10, the most powerful, versatile, and flexible model Google has built to date is officially released.
The model is divided into three different versions according to different scenarios:
Gemini Ultra: Google's largest and most powerful model for highly complex tasks.
Gemini Pro: the best model for all kinds of tasks.
Gemini Nano: the most efficient model for devices such as mobile phones.
According to the benchmark results posted by Google, Gemini shows very strong performance compared with OpenAI's GPT-4, and Gemini is ahead of GPT-4 in other benchmarks except for the Helaswag dataset.
Of the 32 widely used academic benchmarks, 30 of the Gemini Ultra surpassed its current leading level, with a score of 90A score of 0% became the first model to outperform a human expert on the MMLU (Massive Multitasking Language Understanding) test, which combines 57 subjects such as mathematics, physics, history, law, medicine, and ethics.
It also achieved a 59 in the MMMU benchmarkWith a score of 4%, the test covers multiple domains and consists of multimodal tasks that require careful reasoning.
According to Google, Gemini 10 is trained to recognize and understand text, images, audio, etc., so it is better able to understand nuanced information, answer questions related to complex topics, and is especially good at explaining reasoning in complex subjects such as mathematics and physics.
Bard will use a fine-tuned version of Gemini Pro for more advanced reasoning, planning, and comprehension, and will be available in English in more than 170 countries and territories, with plans to expand different modalities in the future and support new languages and regions. In the coming months, Gemini will be applied to Google Search, Ads, Chrome, and Duet AI.
In addition, starting December 13, developers and enterprise customers will be able to access Gemini Pro's Gemini API through Google AI Studio or Google Cloud Vertex AI.