Summary

  • Gemini, Google's new AI model, is the company's "largest and most capable" and may position Google ahead of its competitors in various ways.
  • It is suited for cloud servers but also has a Nano version for devices like Google Pixel 8 Pro, with future availability on Android devices running Qualcomm, MediaTek, and Samsung silicon.
  • Gemini outperforms GPT-4 in many benchmarks while also surpassing human performance, marking a significant advancement in language understanding.

The AI race is still ongoing, with OpenAI, Microsoft, and Google all trading blows in introducing new features that aim to leapfrog the competition. Now, Google may have taken its biggest leap yet with the launch of Gemini. An AI model that the company is calling its "largest and most capable" may well push Google ahead of the competition in multiple ways. It appears to be more suited to running on cloud servers, but the company has also launched a Nano version that will be able to run on devices like the Google Pixel 8 Pro. Google also said in a blog post that it will be available in the future on any of the best Android devices running Qualcomm, MediaTek, and Samsung silicon.

Google Bard is using a tuned version of Gemini Pro starting today, which should allow for it to have advanced reasoning, planning, and understanding. On top of that, it will also be usable in Google Search in the future to, as Google says, enhance generative experiences. There will also be a Bard Advanced experience early next year, though it's unclear what that is, aside from the fact that it will include Gemini Ultra. The company will be launching a "trusted tester program soon" to see how Bard Advanced fares.

Table of scores from Google's Gemini AI
Source: Google

What's especially interesting about Gemini is that Google says it's capable of outperforming GPT-4 in most benchmarks, and it outperforms humans for the first time in MMLU, or massive multitask language understanding.

Developers will be able to make use of Gemini in their applications through Google's AI Studio and Google Cloud Vertex AI starting December 13th. For now, you'll have to play around with the Pro version in Google Bard to see how good it is. While it can achieve record scores in various benchmarks, those pertain to the Ultra version, and it remains to be seen just how powerful the Pro version that most people will be able to use really is. Users in Europe will also have to wait before they can use Gemini Pro.