Google on Tuesday announced the release of Gemini 3, the latest version of its flagship large language model, which it calls its smartest model yet, with plenty of benchmarks to support it. The company said Gemini 3 Pro will be available to users in preview across multiple Google products starting today, including being deployed in Search, and the “advanced reasoning model” Gemini 3 DeepThink will roll out to Google AI Ultra customers after security testing is completed.
If you follow any AI-obsessed types on social media, you know that Gemini 3 has been wildly anticipated by a group that continues to whisper about how the model will be a game-changer. Time will tell whether it actually achieves this, but Google has plenty of metrics to show how the model is better on paper than previous iterations. The company claimed it now tops the leaderboard of LMArena, a benchmarking tool used to compare LLMs, having overtaken Grok 4.1 Thinking. Google also claims that the Gemini 3 Pro demonstrates “PhD-level reasoning” on the Humanities Last Exam and GPQA Diamond, and sets a new record for math performance on MathArena Apex. Of course, the score goes even higher with Gemini 3 Deep Think.
Does it matter that AI benchmarks are considered unreliable and misleading? If you’re on the hype train, not so.
So the Gemini 3 is smarter, which is the primary upgrade brought by the new LLM model. But Google is also introducing some new capabilities with this model. The company announced Google AntiGravity, an integrated development environment (IDE) for coding, which has a built-in AI agent to maximize your vibe coding. Basically, it’s a way to more easily delegate coding tasks to AI, which can work across editors, terminals, and browsers based on user commands. While Gemini 3 Pro will be the default, the company said AntiGravity Cloud also supports SONET 4.5 and GPT-OSS agents. AntiGravity will be available for testing on Windows, Mac, and Linux starting today.
In addition to the newly launched AntiGravity, developers will be able to access Google’s Vertex AI and AI Studio on Gemini 3 Pro. The company also claims that Gemini 3’s coding benchmarks are world-beaters – topping the WebDev Arena leaderboard and setting a new high on Terminal-Bench 2.0. Again, your mileage may vary as to whether they mean anything to you or they are just test numbers.
The great thing about Gemini 3 is that you won’t be able to escape from it. For the first time, the company is integrating models into instant search. So whenever you get a reply from the AI mode, it will go through Gemini 3.
The Gemini app is where the models will primarily reside, and users will get what Google calls a “generative interface”, which offers two output modes called Visual Layout and Dynamic View. Visual layout is a more traditional experience, drawing images to the output based on user prompts. Dynamic View lets Gemini quickly create a website-style interface with functional buttons and lets you interact with different pages of information.
Google’s Gemini 3 is the latest model from one of the biggest players in the AI field to join the game of oneupsmanship. Earlier this month, OpenAI removed its GPT-5.1 model, and on Monday, Elon Musk’s xAI released Grok 4.1. It took full 24 hours for Google to come into the limelight. We’ll see how long the company can keep it before another model usurps it.