Google on Tuesday announced Gemini 3, calling it another big step on the path toward AGI.
“It’s state-of-the-art in reasoning, built to grasp depth and nuance — whether it’s perceiving the subtle clues in a creative idea, or peeling apart the overlapping layers of a difficult problem,” said Google CEO Sundar Pichai in a statement.
Google said it is rolling out Gemini 3 across its major products, including Search. The model is now live in AI Mode in Search with expanded reasoning capabilities and new dynamic experiences.
The model is also available in the Gemini app, as well as to developers through AI Studio, Vertex AI and Google’s new agent-focused development platform, Google Antigravity.
Demis Hassabis, CEO of Google DeepMind, and Koray Kavukcuoglu, the company’s CTO and chief AI architect, announced in a joint statement that Gemini 3 Pro is now available in preview.
“We’re beginning the Gemini 3 era,” they said, noting that the model is being integrated into Search, Workspace, the Gemini app and developer platforms.
Google said Gemini 3 Pro outperforms Gemini 2.5 Pro, OpenAI GPT-5.1 and Claude Sonnet 4.5 across major AI benchmarks, including LMArena, Humanity’s Last Exam, GPQA Diamond and MathArena Apex.
The company highlighted improvements in multimodal capabilities, citing scores of 81% on MMMU-Pro and 87.6% on Video-MMMU. It also recorded 72.1% on SimpleQA Verified, a measure of factual accuracy.
The launch also introduced Gemini 3 Deep Think, an improved reasoning mode. Google said it scores 41% on Humanity’s Last Exam, 93.8% on GPQA Diamond and 45.1% on ARC-AGI-2 with code execution. “Deep Think pushes the boundaries of intelligence even further,” the company said.
With broader multimodal input, longer context and new planning abilities, Google said users can apply Gemini 3 to tasks such as analysing research papers, translating handwritten family recipes, generating visualisations, or evaluating sports performance. In Search, AI Mode now supports generative UI elements and interactive simulations.
For developers, Google launched Google Antigravity, an agent-first development platform built around Gemini 3. The company said Antigravity allows agents to “autonomously plan and execute complex, end-to-end software tasks” with direct access to an editor, terminal and browser. Gemini 3 also integrates with tools including Google AI Studio, Vertex AI, Gemini CLI, Cursor, GitHub, JetBrains and Replit.
The model’s long-horizon planning was cited as another improvement. Google said Gemini 3 Pro leads the Vending-Bench 2 leaderboard, sustaining consistent decision-making over a simulated year of operations.
Subscribers to Google AI Ultra can access these agentic capabilities through Gemini Agent in the Gemini app.
Google emphasised expanded safety testing, saying Gemini 3 has undergone its most extensive evaluations to date, including assessments by external partners such as Apollo, Vaultis and Dreadnode.
“Gemini 3 is our most secure model yet,” the company said, noting reduced sycophancy, better prompt-injection resistance and stronger protection against misuse.
ALSO READ: EU Data Act Goes Live—Why Today Marks a Turning Point for Enterprise Strategy
