Google Unveils Gemini, Its Most Strong AI Model

Google introduced Gemini, a new multimodal general AI (artificial intelligence) model, to rival offerings from Microsoft, OpenAI, and Meta.

Google claims that Gemini, their “largest and most capable” large language model, excels in performance against numerous top benchmarks. Created by Google DeepMind AI, this versatile AI model is powered by Google’s Tensor Processing Units (TPU) for unparalleled speed compared to previous, less advanced models. Gemini can effortlessly understand, process, and integrate various forms of data, including text, code, audio, images, and video.

The company introduces three distinct variants of the Gemini 1.0: the highly capable Gemini Ultra for complex tasks; the scalable Gemini Pro suitable for a wide array of tasks; and the efficient Gemini Nano designed for on-device operations.

Sundar Pichai, CEO of Google and Alphabet, marks these models as the first in the promising new era of Gemini and considers them one of the company’s most significant scientific and engineering endeavors.

Google DeepMind reveals that Gemini Ultra outperforms GPT-4 in 30 out of 32 widely adopted academic benchmark tests featuring areas like image understanding and mathematical reasoning.

Of particular note, the company highlights Gemini Ultra’s 90% score on the MMLU (massive multitask language understanding) test, which encompasses 57 diverse subjects like mathematics, physics, history, law, medicine, and ethics. This achievement marks it as the first AI model to surpass human experts on this specific benchmark.

Additionally, Google announces that Gemini Ultra earned a score of 59.4% on the new MMMU benchmark comprising multimodal tasks across different domains requiring purposeful reasoning. Remarkably, it surpassed previous models in image benchmarks without help from optical character recognition (OCR) systems typically needed for further processing.

Availability of Gemini AI

Google states that the Pro version is currently accessible through Bard chatbot and will soon expand to over 170 countries and territories with plans to accommodate different modalities and support new languages and locations. Beginning December 13, developers and enterprise customers can access Gemini Pro via the Gemini API in Google AI Studio or Google Cloud Vertex AI.

Additionally, in the upcoming months, Google intends to gradually incorporate Gemini into its Search, Ads, Chrome, and other services. Gemini Nano will also be available on Pixel 8 Pro smartphones. Furthermore, Android developers will be able to utilize Gemini Nano through AI Core, a new system feature that will be accessible in Android 14, first on Pixel 8 Pro smartphones. This will be an early preview.

In conclusion, the most sophisticated version of this AI model, Gemini Ultra, is slated for release via Bard Advanced in early 2024. Google will initially provide access to a select group of customers, developers, partners, and safety and responsibility experts for “early experimentation and feedback” prior to a broader rollout to developers and enterprise clients next year.