Google Gemini Overview

---

---

What is Google's Gemini?

Gemini is a new and powerful AI model recently announced and made available from Google. Gemini is now powering Bard, Google's AI-powered chatbot that competes with OpenAI's ChatGPT, Anthropic's Claude, Meta's LLaMa, xAI's Grok, and others.

According to Google, Gemini Ultra is able to beat GPT-4 in several different categories, but the benchmarks used are being debated througout the industry.

The strength of Gemini is its ability to work across modalities. This means that it can simultaneously process and respond to text, images, video, audio, code, and other forms of input.

The model comes in three sizes, "Ultra" is its largest and most powerful model, and can also take the most time to complete complex tasks. "Pro" is its standard model built for speed and efficiency. "Nano" has a smaller footprint (fewer parameters) and is built for use on mobile devices.

Key Statistics

Parameters: Gemini Ultra is estimated to have been built with 1.56 trillion parameters

Context Length: 32,768 tokens

Machine Learning Model: Decoder-only transformers

Cost to Train: Estimated $1 billion (wide variance)

Links:

Google Gemini Intro Video

Google Gemini Demo Video

Google Gemini Page

API Documentation