A few weeks short of Gemini 2’s first birthday, Google has announced Gemini 3 Pro. Naturally, the company claims the new system is its most intelligent AI model yet, offering state-of-the-art reasoning, class-leading vibe coding performance and more. The good news is you can put those claims to the test today, with Google making Gemini 3 Pro available across many of its products and services.
Google is highlighting a couple of benchmarks to tout Gemini 3 Pro’s performance. In Humanity’s Last Exam, widely considered one of the toughest tests AI labs can put their systems through, the model delivered a new top accuracy score of 37.5 percent, beating the previous leader, Grok 4, by an impressive 12.1 percentage points. Notably, it achieved its score without turning to tools like web search. On LMArena, meanwhile, Gemini 3 Pro is now on top of the site’s leaderboards with a score of 1,501 points.
Okay, but what about the practical benefits of Gemini 3 Pro? In the Gemini app, the new model will translate to answers that are more concise and better formatted. It also enables a new feature Google calls Gemini Agent. The tool builds on Project Mariner, the web-surfing Chrome AI the company debuted at the end of last year. It allows users to ask Gemini to complete tasks for them. For example, say you want help managing your email inbox. In the past, Gemini would have offered some general tips. Now, it can do that work for you.
To try Gemini 3 Pro inside of the Gemini app, select “Thinking” from the model picker. The new model is available to everyone, though AI Plus, Pro and Ultra subscribers can use it more often before hitting their rate limit. To make the most of Gemini Agent, you’ll need to grant the tool access to your Google apps.
In Search, meanwhile, Gemini 3 Pro will debut inside of AI Mode, with availability of the new model first rolling out to AI Pro and Ultra subscribers. Google will also bring the model to AI Overviews, where it will be used to answer the most difficult questions people ask of its search engine. In the coming weeks, Google plans to roll out a new routing algorithm for both AI Mode and AI Overviews that will know when to put questions through Gemini 3 Pro. In the meantime, subscribers can try the new model inside of AI Mode by selecting “Thinking” from the dropdown menu.
In practice, Google says Gemini 3 Pro will result in AI Mode finding more credible and relevant content related to your questions. This is thanks to how the new model augments the fan-out technique that powers AI Mode. The tool will perform even more searches than before and with its new intelligence, Google suggests it may even uncover content previous models may have missed. At the same time, Gemini 3’s better multi-modal understanding will translate to AI Mode generating more dynamic and interactive interfaces to answer your questions. For example, if you’re researching mortgage loans, the tool can create a loan calculator directly inside of its response.
For developers and its enterprise customers, Google is bringing Gemini 3 to all the usual places one can find its models, including inside of the Gemini API, AI Studio and Vertex AI. The company is also releasing a new agentic coding app called Antigravity. It can autonomously program while creating tasks for itself and providing progress reports. Alongside Gemini 3 Pro, Google is introducing Gemini 3 Deep Think. The enhanced reasoning mode will be available to safety testers before it rolls out to AI Ultra subscribers.
This article originally appeared on Engadget at https://www.engadget.com/ai/googles-new-gemini-3-model-arrives-in-ai-mode-and-the-gemini-app-160054273.html?src=rss
