On February 8th, Google released Gemini Ultra for use by the general public. This is Google’s major move into the AI industry, taking a strike at fellow giants like OpenAI, Microsoft, and Meta, creators of ChatGPT, Copilot, and the LLaMA family respectively.

This was alongside a rebranding of Google’s Bard AI into Gemini to highlight this advancement, with the Gemini service branching into two versions: Gemini and Gemini Advanced. The former provides access to Gemini Pro, the intermediate of the three model types of Gemini, and the latter provides access to Gemini Ultra, of which is only available to use if subscribed to the newly introduced paid plan, named the Google One AI Premium plan. As of February 15th, a 2-month free trial is being offered, with a $19.99 charge beyond that.

Google also announced plans to build integrations between Gemini and their suite of productivity services, like Google Docs, Gmail, Sheets, and more, which should give a seamless experience between them.

This marks a milestone in Google’s intrepid journey of getting ahead in the quickly-evolving AI race, and gaining control of the market that surrounds it. Gemini Ultra’s public distribution is preceded by a long line of research and development, from the AI that correctly interprets mistyped queries to Gemini Ultra, something Google is hoping could supersede ChatGPT. Specifically, Gemini Ultra is being pitted against OpenAI’s GPT-4, a model so advanced that it’s currently acting as the gold standard of AI.

More trouble will soon come to paradise as Google announced Gemini 1.5 on February 15th, the successor to the current Gemini 1.0 family. Gemini 1.5 uses a mixture-of-experts (MoE) structure, something GPT-4 also uses, in which the model is essentially composed of multiple smaller models that specializes in specific areas of interest. This propels performance while also maintaining efficiency, as queries would only need to evoke some of these smaller specialized models rather than one big generalist model. Gemini 1.5  The real kicker is the fact that Gemini 1.5 is able to process up to a million tokens, potentially even 10 million tokens. (Tokens are like chunks of words/data that AI processes in). According a Google blogpost, a million tokens roughly equates to “1 hour of video, 11 hours of audio, codebases with over 30,000 lines of code or over 700,000 words.” GPT-4 can process up to 128,000 tokens.

Those are only but a few details about Gemini 1.5. It’s actually public, though if by public, it meant a limited number of developers selected by a process.

Usage of Gemini Ultra through Gemini Advanced is available at gemini.google.com.