Google launches Gemini 3, its most intelligent model ever

Google has a new contender for the smartest AI in the world.

On Tuesday, the company launched Gemini 3, calling it its “most intelligent model.”

The new LLM (large language model) should have improved reasoning, helping it to understand the depth and nuance of the tasks it tackles. But according to Google, it should also be better at figuring out “the context and intent behind your request,” which means you should be able to get things done with less prompts.

Gemini hasn’t only been updated on its own. Google said it shipped the LLM at scale across its products, meaning that Gemini 3 is already live in AI mode in Google Search — the first time the company has shipped a new model in Search on launch day (note, however, that it’s currently only live for Google AI Pro and Ultra subscribers). Gemini 3 is now also live in the Gemini app, in AI Studio and Vertex AI, as well as in Google’s agentic development platform, Antigravity.

As is customary, Google immediately showed off Gemini 3’s score in various benchmarks. For example, Gemini 3 Pro is now the leader in LMArena with a score of 1501 Elo, ahead of xAI’s Grok, and Gemini 3’s predecessor, Gemini 2.5 Pro.

And if you want to push it a little further, you can run Gemini 3 Deep Think, which is even better in reasoning and more suited for the most complex tasks. That model is notably better in certain benchmarks, including the notoriously tough Humanity’s Last Exam.

On top of all that, Google also redesigned its Gemini app. One key change is the My Stuff folder, where you can find the various chats and documents you’ve created with Gemini. The company also improved the shopping experience, and started experimenting with generative interfaces, which are created by the model as you prompt it.

Finally, the Gemini Agent is a feature which can complete multi-step tasks for you, including managing your appointments and reminders, organizing your inbox, or performing research online.