Google DeepMind unveils Gemini 3, its most advanced AI model yet

Google DeepMind has released Gemini 3, the latest iteration of its flagship AI model, positioning it as a major step toward artificial general intelligence. The model combines improved reasoning, multimodal capabilities, and agentic features, with CEO Sundar Pichai noting that "Gemini 3 is our most intelligent model, that combines all of Gemini's capabilities together."

State-of-the-art performance across benchmarks#

Gemini 3 Pro significantly outperforms its predecessor Gemini 2.5 Pro across major AI benchmarks:

• Tops the LMArena Leaderboard with a breakthrough score of 1501 Elo
• Demonstrates PhD-level reasoning on Humanity's Last Exam (37.5% without tools) and GPQA Diamond (91.9%)
• Achieves 23.4% on MathArena Apex for mathematical reasoning
• Scores 81% on MMMU-Pro and 87.6% on Video-MMMU for multimodal understanding
• Reaches 72.1% on SimpleQA Verified for factual accuracy

The company also introduced Gemini 3 Deep Think, an enhanced reasoning mode that pushes performance even further, achieving 41.0% on Humanity's Last Exam and 93.8% on GPQA Diamond.

Three core capabilities#

Learn anything — The model leverages a 1 million-token context window to synthesize information across text, images, video, audio and code. Examples include deciphering handwritten recipes in multiple languages, generating interactive guides from academic papers, and analyzing sports performance with training recommendations.

Build anything — Gemini 3 is described as "the best vibe coding and agentic coding model" built yet, scoring 1487 Elo on the WebDev Arena leaderboard and 76.2% on SWE-bench Verified. It's available in Google AI Studio, Vertex AI, and third-party platforms like Cursor, GitHub, and JetBrains.

Plan anything — The model excels at long-horizon planning and multi-step workflows. It tops the Vending-Bench 2 leaderboard, which tests complex decision-making over extended periods. Agentic capabilities are now available to Google AI Ultra subscribers in the Gemini app.

New development platform#

Google is launching Google Antigravity, an agent-first development platform that gives AI agents direct access to the editor, terminal, and browser. Agents can autonomously plan and execute complex software tasks while validating their own code, fundamentally reimagining the developer experience.