Google released Gemini 3 on Tuesday, rolling out what it calls its most advanced AI model across its entire ecosystem. The release also includes a new coding platform called Antigravity, and for the first time, Google integrated its newest Gemini model into Search on launch day.
The timing carries weight. Coming just seven months after Gemini 2.5 and less than a week after OpenAI released GPT 5.1, the rollout underscores the pace at which leading AI companies are advancing their technology. Google is making Gemini 3 immediately available through the Gemini app, which has more than 650 million monthly users, and through developer platforms including AI Studio and Vertex AI.
“With Gemini 3, we’re seeing this massive jump in reasoning,” said Tulsee Doshi, Google’s head of product for the Gemini model. “It’s responding with a level of depth and nuance that we haven’t seen before.”
Gemini 3: What’s actually new
Let’s get nerdy for a second. Here’s a look at Google’s latest LLM by the numbers:
- Gemini 3 Pro scored 1501 Elo on the LMArena leaderboard, topping virtually every other LLM, including Claude, ChatGPT, and Grok.
- On the GPQA Diamond benchmark, which tests PhD-level scientific reasoning, it achieved 91.9%—better than Claude Sonnet 4.5 and ChatGPT 5.1.
- The model also scored 37.5% on Humanity’s Last Exam without tools, surpassing GPT-5 Pro’s previous high of 31.64%.
- In math, Gemini 3 set a new standard with 23.4% on MathArena Apex.
What distinguishes this release is Google’s emphasis on agentic capabilities—the model’s ability to plan and execute multi-step tasks with reduced human intervention. Demis Hassabis, CEO of Google DeepMind, described Gemini 3 as evolving from “simply reading text and images to reading the room.” The model combines what Google calls state-of-the-art reasoning with multimodal understanding, processing text, images, video, audio and code simultaneously.
That multimodal strength shows up in benchmarks focused on visual and video understanding. Gemini 3 scored 81% on MMMU-Pro and 87.6% on Video-MMMU—higher marks than Claude Sonnet 4.5 and ChatGPT 5.1.
The model can also generate what Google calls “generative UI”—designing custom interfaces in real-time based on prompts, from interactive physics simulations to mortgage calculators.
What is Google Antigravity?
Antigravity represents Google’s effort to reimagine developer tools for this new generation of AI.
Unlike previous setups where an AI chatbot sits in the corner of your screen and you ask it questions, Antigravity puts the AI in charge of a dedicated workspace. The AI can look at your code, understand what you’re trying to build, write code itself, test it, and catch problems—all with less human intervention needed. Rather than treating AI as an assistant within the editor, Antigravity elevates agents to a dedicated surface with direct access to the editor, terminal, and browser.
“Using Gemini 3’s advanced reasoning, tool use and agentic coding capabilities, Google Antigravity transforms AI assistance from a tool in a developer’s toolkit into an active partner,” Google said in its announcement. The platform allows agents to autonomously plan and execute software tasks while validating their own code.
Antigravity is free during its public preview, though users have reported rate limits that refresh every five hours. The platform includes access to Gemini 3 Pro, Anthropic’s Claude Sonnet 4.5, and OpenAI’s GPT-OSS.
Stitching Gemini and Search together
The Search integration marks another departure from past practice. “This is the first time we’re shipping Gemini in Search on day one,” Alphabet CEO Sundar Pichai wrote in the company’s announcement post.
Paying subscribers to Google AI Pro ($20 per month) and Ultra ($250 per month) can access Gemini 3 in Search’s AI Mode, which uses the model’s reasoning capabilities to generate visual layouts with interactive elements.
“Gemini 3 is also making Search smarter by re-architecting what a helpful response looks like,” Stein said. “With new generative UI capabilities, Gemini 3 in AI Mode can now dynamically create the overall response layout when it responds to your query—completely on the fly.”
Google is also releasing Gemini 3 Deep Think, an enhanced reasoning mode that further improves performance on complex problems. The mode achieved 41.0% on Humanity’s Last Exam without tools and 93.8% on GPQA Diamond—as you might imagine, both were best-in-class. Deep Think will become available to Google AI Ultra subscribers after additional safety testing.
Major development platforms are integrating the model. GitHub reported that Gemini 3 Pro demonstrated 35% higher accuracy in resolving software engineering challenges than Gemini 2.5 Pro in early testing. JetBrains noted more than a 50% improvement in the number of solved benchmark tasks. The model is also being added to Cursor, Manus, Replit and other coding tools. There’s a lot of positivesentiment online right now, but of course, it’s still early days.
The release aims to put Google back in a competitive position it lost when ChatGPT launched in late 2022. After facing criticism over early Gemini outputs and setbacks with AI Overviews in Search, the company has steadied its approach. AI Overviews now reach 2 billion users monthly, and more than 70% of Google Cloud customers use the company’s AI products.
For this story, Coins2Day used generative AI to help with an initial draft. An editor verified the accuracy of the information before publishing.