Google just dropped Gemini 3, and it’s already shattering benchmarks. Gemini 3 Pro now sits at the top of the LMArena leaderboard with what Google calls a “breakthrough score,” trouncing Gemini 2.5 Pro across every major test — math, long-form reasoning, multimedia understanding, you name it. Oh, and yes, it translates, too. This is the smartest model Google has built, period.
The good news is Google isn’t wasting any time rolling it out, either. Gemini 3 is already live in Google Search, the Gemini app and a full suite of developer tools — marking one of the company’s most aggressive moves yet in the AI arms race.
Starting today, Google AI Pro and Ultra subscribers in the U.S. can start using Gemini 3 Pro directly inside Search by selecting “Thinking” from the model selector.
My first impressions
I put Gemini 3 through its paces this morning, and it’s clear Google’s latest model is so much faster and smarter. It's evident from the first prompt you try. The new “Thinking” mode inside Google Search answered layered questions with contextual depth I haven’t seen from previous versions. It connected dots across multiple queries, synthesized background info and pulled in relevant sources I didn’t explicitly mention.
In Gemini Advanced, I tested code generation and complex multi-step prompts, and the improvement in reasoning was obvious. Where Gemini 2.5 would occasionally hallucinate or misinterpret my intent, Gemini 3 nailed the logic.
Contextual awareness was especially clear in math, writing and code tasks; essentially everything I use Gemini for, making the switch to Gemini 3 a no-brainer. It’s also noticeably better at following nuanced instructions. For the first time, I felt like Gemini could rival ChatGPT-5.1 and Claude Sonnet when it comes to understanding, not just answering. Keep an eye out for my full face off of ChatGPT-5 and Gemini 3.0 coming soon.
Google Search gets smarter (and more visual)
Google's also reworked how Search finds information behind the scenes; a complete game-changer for the way we Google things. Gemini 3 digs deeper and genuinely understands what you're actually asking for in terms of context, not just keywords.
Gemini 3’s deeper reasoning allows Search to perform more background queries, uncover sources older models would have missed and better understand the intent behind your question, beyond the keywords. The result is cleaner, more accurate answers with fewer irrelevant results or hallucinations.
And there’s more on the way. Google says automatic model routing is coming soon, sending simple questions to lighter models and reserving Gemini 3 for the hard stuff. It is efficiency meets intelligence — and a clear sign of where Search is headed next.
Search now builds custom tools for your question
This is where things get really interesting. Instead of just spitting out text answers, Gemini 3 can now generate interactive tools, visualizations and even simulations right inside your search results.
Curious about the physics of the three-body problem? Gemini 3 will build you a live simulation you can actually play with. Shopping for a mortgage? It'll generate an interactive loan calculator where you can plug in different numbers and see the results instantly. The layout adapts to your question so no two searches look quite the same.
What Makes Gemini 3 Different
Google is calling Gemini 3 its "most intelligent model" yet, and the specs back that up:
- It crushes the benchmarks. Gemini 3 Pro tops the LMArena leaderboard and beats Gemini 2.5 across the board — math, long-form problem solving, understanding images and videos and more.
- Deep Think mode is coming. For the really hard problems, Google's rolling out an enhanced reasoning mode called Deep Think. It's built for long-term planning and novel problem-solving. Google AI Ultra subscribers will get this first.
- It can do your work for you. The new Gemini Agent can tackle multi-step tasks on your behalf — things like organizing your inbox, automating workflows, or planning trips. This rolls out to Ultra subscribers first.
- Google promises that it is safer. Gemini 3 has stronger defenses against prompt injection attacks and won't just agree with you for the sake of it. Google's been serious about making this its most secure model yet.
Soon, Google will also roll out automatic model selection. Simple searches questions will get answered faster with lighter models, while the tough stuff gets routed to Gemini 3.
And, Google says this is just the beginning and users can expect more dynamic visual tools and creative layouts to roll out in the upcoming months.
The bottom line
Gemini 3 is Google's biggest swing at AI yet. It's smarter, more interactive and can actually handle even bigger complex tasks. But the real shift is how tightly Google's weaving it into Search itself.
With visual tools that build themselves, simulations that run in real time and deeper reasoning happening behind the scenes, Search is completely shifting from a search engine towards an AI assistant that actually knows what you're trying to do.
Gemini 3 is already available within the Gemini app and AI Mode in Search Google AI Pro and Ultra users in the US only for now. Deep Think mode launches in the coming weeks for Ultra subscribers.
Follow Tom's Guide on Google News and add us as a preferred source to get our up-to-date news, analysis, and reviews in your feeds.
More from Tom's Guide
- 5 free Google AI tools that blew me away — and most people don’t know they exist
- I gave ChatGPT, Claude and Gemini $50 to improve my life — their ideas surprised me
- This AI sticker toy has kids ditching screen time — and it's selling out fast
You must confirm your public display name before commenting
Please logout and then login again, you will then be prompted to enter your display name.
