Google launched Gemini 1.0, its next-gen multimodal AI model surpassing OpenAI's GPT-4

Google has finally launched Gemini 1.0, it's next-generation AI model. Available through Google Gemini, Gemini is designed to comprehend and operate across multiple modes, including text, code, audio, images, and video. Google has chosen multimodal training for Gemini from the beginning, using TPU 4 and TPU v5e, to enhance its understanding, reasoning, and coding capabilities for complex tasks.

Gemini's abilities were showcased through its processing and summarization of 200,000 scientific research papers within an hour and its capacity to understand, explain, and generate code in various programming languages. Gemini 1.0 is available in three sizes: Gemini Ultra for complex tasks, Gemini Pro for scaling across tasks, and Gemini Nano for on-device tasks.

Gemini Ultra outperformed OpenAI's GPT-4 (the main model used for ChatGPT) in text-based tasks in performance benchmarks and was the first model to surpass human experts on MMLU at 90.0%. It also outperformed GPT-4V in multimodal benchmarks, showing signs of complex reasoning abilities. Gemini underwent extensive safety evaluations and has protections to address multimodal capabilities and bias.

As we mentioned at the beggining, Gemini Pro has been integrated into Google Gemini (a 'fine-tuned version') and launched in English for 170 countries, with UK and European availability coming soon, and Gemini Ultra will be released next year after extensive safety checks and model refinements. Gemini is expected to be integrated with Google Search, Google Chrome, Google Duet AI, and Google AdSense in the coming months.