Google Unveils Gemini 3: A Leap Forward In AI Intelligence And Capabilities
Google has officially launched Gemini 3, its most advanced artificial intelligence model to date, promising a significant upgrade in reasoning, multimodal understanding, and overall helpfulness across its product ecosystem. The new model, which arrives just shy of the first anniversary of Gemini 2, is being integrated into key Google services, including Search, the Gemini app, and developer platforms like AI Studio and Vertex AI. Google asserts that Gemini 3 represents a substantial leap forward, offering state-of-the-art reasoning and class-leading performance in various benchmarks.
One of the most notable achievements highlighted by Google is Gemini 3 Pro’s performance on “Humanity’s Last Exam,” a rigorous test for AI systems. The model achieved a new top accuracy score of 37.5%, surpassing its predecessor and competitors by a significant margin without the use of external tools like web search. Furthermore, Gemini 3 Pro has claimed the top spot on the LMArena leaderboard with a score of 1,501 points, underscoring its enhanced capabilities in complex problem-solving and information processing.
The Gemini app is set to receive a major overhaul with Gemini 3, featuring more concise and better-formatted responses. A key new functionality is “Gemini Agent,” an advanced tool built upon Google’s Project Mariner. This agent allows Gemini to perform tasks on behalf of the user, such as managing email inboxes or planning complex trips, moving beyond providing general advice to actively executing tasks. For developers, Google is introducing “Google Antigravity,” a new agentic development platform designed to enable the creation of AI-powered agents that can autonomously plan, code, and validate software tasks.
In Google Search, Gemini 3 will debut within “AI Mode,” aiming to provide more credible and relevant content for user queries. The model’s improved multimodal understanding is expected to lead to more dynamic and interactive interfaces within Search results. Google is also rolling out a new routing algorithm to optimize when questions are directed to Gemini 3, ensuring users benefit from its advanced reasoning for the most challenging queries.
Beyond Gemini 3 Pro, Google is also introducing “Gemini 3 Deep Think,” an enhanced reasoning mode designed for even more complex problems. This mode is currently available to safety testers and will soon be rolled out to Google AI Ultra subscribers. In testing, Gemini 3 Deep Think has demonstrated superior performance on benchmarks like Humanity’s Last Exam and GPQA Diamond, showcasing its potential to tackle novel and highly challenging AI tasks, pushing the boundaries of artificial intelligence closer to human-like intelligence.