Google has officially unveiled Gemini 3, its most advanced AI model yet, marking a major leap in reasoning, multimodal understanding and agentic capabilities. The model is rolling out across Search, the Gemini app, AI Studio, Vertex AI and Google’s new development platform, Antigravity.
Gemini 3: All Details
Gemini 3 is designed to understand deeper context, interpret complex requests with less prompting and deliver more accurate and insightful responses. It improves significantly on Gemini 2.5 with stronger reasoning, better intent detection and more natural interaction across text, images, video, audio and code.
Google is also shipping Gemini 3 directly into AI Mode in Search on day one — a first for the company — enabling more dynamic, generative results and richer UI elements that adapt to user queries.
The first model in the series, Gemini 3 Pro, launches today in preview. It outperforms Gemini 2.5 Pro across every major benchmark, topping the LMArena leaderboard with a score of 1501 Elo. It shows major gains in scientific and mathematical reasoning, scoring 37.5% on Humanity’s Last Exam without tools, 91.9% on GPQA Diamond and a new record of 23.4% on MathArena Apex.
Its multimodal performance also improves sharply, hitting 81% on MMMU-Pro, 87.6% on Video-MMMU and 72.1% on SimpleQA Verified, signalling a stronger grip on factual accuracy and complex visual reasoning.
Gemini 3 Pro also brings a new level of depth and nuance to every interaction. “Its responses are smart, concise and direct, trading cliché and flattery for genuine insight — telling you what you need to hear, not just what you want to hear. It acts as a true thought partner that gives you new ways to understand information and express yourself, from translating dense scientific concepts by generating code for high-fidelity visualisations to creative brainstorming,” said Google.
Google also announced Gemini 3 Deep Think, an enhanced reasoning mode that surpasses Gemini 3 Pro on several advanced benchmarks. It reaches 41% on Humanity’s Last Exam, 93.8% on GPQA Diamond and an unprecedented 45.1% on ARC-AGI-2 with code execution. Deep Think will be available to Google AI Ultra subscribers after additional safety reviews.
Read More: Google Rolls Out Shopping Features in Gemini Alongside Other Agentic AI Shopping Experiences
Gemini 3 further introduces richer multimodal learning tools, such as translating handwritten recipes, generating interactive study materials from research papers or analysing sports videos to offer personalised coaching. In Search, Gemini 3 now powers new generative UI experiences that create diagrams, layouts and interactive simulations in real time.
Developers can start building with Gemini 3 today using AI Studio, Vertex AI, the Gemini CLI and Google Antigravity. The model delivers more autonomous agentic coding, ranks first on the WebDev Arena leaderboard and sets new highs on SWE-bench Verified and Terminal-Bench 2.0.
Gemini 3 also improves long-horizon planning, topping the Vending-Bench 2 leaderboard and demonstrating consistent decision-making over extended simulated workflows. These improvements enable real-world tasks like email organization, booking services and executing multi-step workflows within the Gemini app. The Google AI Ultra plan subscribers can try these agentic capabilities in the Gemini App with Gemini Agent.
Google Antigravity introduces an agent-first development workflow
Alongside the model, Google launched Antigravity, a new agentic development platform where AI agents can plan, write and validate code autonomously. Agents have direct access to the editor, terminal and browser, enabling them to execute end-to-end tasks with minimal user input. Antigravity integrates Gemini 3 Pro, the Gemini 2.5 Computer Use model and Google’s Nano Banana image-editing model.
Safety remains a major focus
Google says Gemini 3 is its most secure model yet, with reduced vulnerability to prompt injections, better resistance to misuse and lower sycophancy. The model underwent extensive internal and third-party safety evaluations, including audits from Apollo, Vaultis and Dreadnode, as well as input from bodies like the UK AISI.
Gemini 3: Availability
Starting today, Gemini 3 begins rolling out in:
• the Gemini app for everyone
• AI Mode in Search for Google AI Pro and Ultra subscribers
• the Gemini API in AI Studio and Vertex AI
• the new Antigravity development platform
• the Gemini CLI
Gemini 3 Deep Think will arrive in the coming weeks after further safety testing. Google also confirmed that additional models in the Gemini 3 family are on the way.