Summary

  • Google has released an experimental version of Gemini 2.0 Advanced with ‘significantly improved performance’ in areas like math, coding, reasoning, and following instructions.
  • The new model is exclusively available to Gemini Advanced subscribers.
  • Gemini 2.0 Advanced is currently limited to Gemini on desktop and mobile web, and it should land on the Gemini Android app soon.



Google announced its latest AI breakthrough in the form of Gemini 2.0 last week. For reference, Google released Gemini 1.0 late last year, back when the AI chatbot was still known as Bard. Fast-forward to May this year, and we got access to Gemini 1.5 Pro, featuring a massive 1 million-token context window. The context window was expanded to 2-million tokens later in June, followed by Gemini being updated to Gemini 1.5 Flash (free) in July.

Earlier this week, the tech giant’s Gemini 2.0 Flash began rolling out on the web and subsequently made its way to the Gemini Android app, complete with a new drop-down model selector. Now, the top-of-the-line Gemini-Exp-1206, which we’ll refer to as ‘Gemini 2.0 Advanced‘ for simplicity’s sake, is beginning its web rollout too, which means it should start landing on Android devices late this week or early next.

Google Gemini

Related

Gemini 2.0 Flash lands on Android with a new model selector in tow

Similar to Gemini on the web

1


The rollout was announced by Google in a blog post, highlighting the new model’s “significantly improved performance” when it comes to solving math problems, coding, reasoning, and instruction following. However, the tech giant also cautioned users ready to pounce that the new model is still in testing, which means that it will continue to lack access to real-time information and compatibility with some Gemini features until it leaves the early experimental phase.

For what it’s worth, the new model, even in its experimental phase, has climbed its way to the #1 spot on the Chatbot Arena LLM Leaderboard, which is essentially an open, community-driven platform for crowdsourced AI benchmarking. Elsewhere, Google has also introduced a new ‘FACTS Grounding‘ benchmark, which essentially evaluates specific models’ accuracy and ability to avoid hallucinations. The benchmarking tool is currently topped by Gemini 2.0 Flash and is yet to be updated with data from Gemini 2.0 Advanced.



Exclusive to Gemini Advanced users, at least for now

The Google Genimi 2.0 logo surrounded by question marks

Related

Gemini 2.0: Five major changes you need to know

Gemini 2.0 brings new AI capabilities to Google’s ecosystem

2

Unlike Gemini 2.0 Flash, access to the top-of-the-line model is strictly limited to Gemini Advanced, which was handed out for free with the purchase of all new Pixel 9 Pro series devices (including the Pixel 9 Pro Fold). If you don’t own a Pixel 9 Pro device, Gemini Advanced is available via the $20/mo Google One AI Premium plan.

If you’re already a subscriber, toggling to Gemini 2.0 Advanced is as simple as tapping on the model drop-down on the top left and selecting ‘2.0 Experimental Advanced.’ The new experimental model is available to try out on desktop and mobile web. If it follows Gemini 2.0 Flash’s pattern, it should land on the Gemini Android app later this week or early next.



A screenshot of Gemini's model drop-down highlighting 2.0 Experimental Advanced.