Google has released Gemini 3, its most advanced artificial intelligence model to date, marking a significant leap in AI reasoning capabilities and multimodal understanding . The launch, announced Wednesday by CEO Sundar Pichai and Google DeepMind leadership, introduces state-of-the-art performance across major AI benchmarks while expanding availability across Google's product ecosystem and developer platforms .
Performance Breakthroughs
Gemini 3 Pro achieves unprecedented scores on industry-standard benchmarks, topping the LMArena Leaderboard with a breakthrough score of 1,501 Elo . The model demonstrates PhD-level reasoning with a 37.5% score on Humanity's Last Exam without tool assistance and 91.9% on GPQA Diamond, while setting new standards in mathematics with 23.4% on MathArena Apex .
The model's multimodal reasoning capabilities show significant advancement, scoring 81% on MMMU-Pro and 87.6% on Video-MMMU, while achieving 72.1% on SimpleQA Verified for improved factual accuracy . These benchmarks indicate Gemini 3 Pro can solve complex problems across diverse topics including science and mathematics with high reliability .
Enhanced Reasoning Mode
Google is introducing Gemini 3 Deep Think, an enhanced reasoning mode that pushes performance boundaries further . In testing, Deep Think achieves 41% on Humanity's Last Exam and 93.8% on GPQA Diamond, while reaching an unprecedented 45.1% on ARC-AGI-2 with code execution . The Deep Think mode is currently undergoing safety evaluations and will be available to Google AI Ultra subscribers in coming weeks .
Developer and Enterprise Tools
Gemini 3 is now available to developers through Google AI Studio, Vertex AI, and the new Google Antigravity platform . The model scores 1,487 Elo on the WebDev Arena leaderboard and achieves 54.2% on Terminal-Bench 2.0, which tests computer operation via terminal, while greatly outperforming Gemini 2.5 Pro on SWE-bench Verified at 76.2% .
Google Antigravity represents a new agentic development platform that transforms AI assistance from a toolkit component into an active development partner . The platform enables autonomous planning and execution of complex software tasks, with agents having direct access to editor, terminal, and browser environments .
Product Integration
Starting today, Gemini 3 is rolling out across Google's product lineup, including the Gemini app with over 650 million monthly users, AI Mode in Search serving 2 billion monthly users, and enterprise solutions through Gemini Enterprise . This marks the first time Google is shipping a new Gemini model in Search on launch day .
The model powers new generative UI experiences in AI Mode in Search, creating immersive visual layouts and interactive tools dynamically based on user queries . For Google AI Ultra subscribers, Gemini Agent capabilities enable the model to navigate complex multi-step workflows, from booking local services to organizing email inboxes .
Evolution from Previous Generations
Gemini 3 builds on the foundation established by previous iterations, combining Gemini 1's native multimodality and long context windows with Gemini 2's agentic capabilities and advanced reasoning . The model features a 1 million-token context window and enhanced spatial understanding, enabling it to process vast amounts of information across text, images, video, audio, and code .
According to Pichai, Gemini 2.5 Pro topped LMArena for over six months, while the broader Gemini ecosystem now serves 70% of Google Cloud customers and supports 13 million developers building with generative models .
Safety and Security
Gemini 3 has undergone the most comprehensive safety evaluations of any Google AI model to date, showing reduced sycophancy, increased resistance to prompt injections, and improved protection against cyberattack misuse . Google partnered with organizations including the UK AISI and independent assessors like Apollo, Vaultis, and Dreadnode for evaluation .
The model's development follows Google's Frontier Safety Framework, with testing conducted by world-leading subject matter experts before public release . Additional models in the Gemini 3 series are planned for release to expand AI capabilities further .
