What you need to know
- Gemini 3 Pro reclaims AI benchmarks, outperforming previous models and dominating both LMArena and WebDev Arena leaderboards.
- New AI mode enhances Google Search with multimodal responses, offering smarter, concise interactions for users.
- Gemini 3 introduces breakthrough features in Google Antigravity, enabling advanced agentic development and independent coding capabilities.
In a significant advancement for artificial intelligence, Google has launched Gemini 3, its most sophisticated AI model to date, now accessible across multiple platforms. With exceptional scores on key benchmarks, Gemini 3 Pro is set to transform user interactions through enhanced reasoning and multimodal capabilities.
This rollout marks a pivotal moment as Google introduces a new AI model in Search from day one. Users of the AI Mode in Search can now tap into the model’s superior reasoning and coding skills to tackle complex queries, while also generating multimodal responses that blend text with interactive visuals. Additionally, the Gemini app features a Gemini Agent capable of managing multi-step tasks autonomously.
The previous Gemini 2.5 models, which debuted in March 2025, had performed admirably on leading AI benchmarks for several months but recently lost traction to competitors like xAI and Anthropic. In contrast, Gemini 3 Pro has swiftly reclaimed its position, leading both the LMArena and WebDev Arena leaderboards immediately after launch. When compared to its predecessor, Gemini 2.5 Pro, the new model demonstrates marked improvements across all major AI benchmarks.
Gemini 3 Pro has achieved a remarkable score of 1,501 points on the LMArena leaderboard, surpassing Grok 4.1 Thinking, which scored 1,483 points. For context, Gemini 2.5 Pro previously held third place with a score of 1,452 points. In other assessments, such as Humanity’s Last Exam, Gemini 3 Pro scored an impressive 37.5% without tool use, showcasing PhD-level reasoning capabilities, as noted by Google.
For everyday users, Gemini 3 is designed to “bring a new level of depth and nuance to every interaction.” Unlike many AI models that tend to cater to user preferences, often providing comforting but inaccurate responses, Gemini 3 Pro aims for “smart, concise, and direct” answers, challenging the perception that AI cannot deliver the truth users need.
For web development and generative user-interface design, Gemini 3 is touted as “the best vibe coding and agentic coding model” Google has ever created. This is evidenced by its leading 1,487 ELO score on the WebDev Arena leaderboard, along with superior performance on Terminal-Bench 2.0 and SWE-bench Verified benchmarks. The capabilities of Gemini 3 Pro extend to building playable 3D video games from scratch and creating websites, catering to diverse coding needs.
To enhance its performance further, Gemini 3 Deep Think has improved results on Humanity’s Last Exam to 41% without tool use, pushing the boundaries of reasoning and multimodal understanding. This model is currently undergoing additional safety evaluations and will be available to safety testers before being rolled out to Google AI Ultra subscribers in the upcoming weeks.
Gemini 3 is now live in Google AI Studio, Vertex AI, Gemini CLI, and Antigravity
Gemini 3 is not only rolling out to Google’s consumer products but is also becoming available for developers. It is now accessible in Google AI Studio, Vertex AI, and Gemini CLI, as well as on third-party platforms like Cursor, GitHub, JetBrains, Manus, and Replit. Most notably, Gemini 3 serves as the foundation for Google Antigravity, a groundbreaking agentic development platform.
Google Antigravity functions as an AI-powered integrated development environment (IDE). While the concept of an IDE is not new, Antigravity distinguishes itself with its intelligent AI agents. The new features of Gemini 3, including advanced reasoning and enhanced agentic coding capabilities, empower AI agents within Antigravity with unprecedented permissions.
These AI agents can directly access the editor, terminal, and browser in Google Antigravity, transforming them into coding partners capable of planning, executing, and validating software tasks independently. The platform utilizes a combination of Gemini 3 Pro, Gemini 2.5 Computer Use, and Nano Banana (Gemini 2.5 Image) to facilitate comprehensive agentic workflows.
Try Gemini 3 now in the Gemini app and Google Search
Gemini 3 Pro is now rolling out to Google’s consumer products, including Search and the Gemini app. Users subscribed to Google AI Pro or Ultra can experience it by selecting the “Thinking” model in AI Mode. While it will soon be available to all AI Mode users in the U.S., it currently remains a paid exclusive.
In the upcoming weeks, AI Pro and Ultra users will benefit from a new automatic model selector in Search, which will direct challenging requests to the flagship Gemini 3 models while routing simpler tasks to quicker, more efficient alternatives.
With its enhanced understanding, Gemini 3 enables AI Mode to utilize the query fan-out method, allowing for multiple simultaneous searches and uncovering new content based on user intent. By accurately interpreting the purpose of a user’s request, Gemini 3 and AI Mode can reveal relevant information that older models might have overlooked.
Furthermore, AI Mode employs generative UI to produce interactive and engaging learning tools in response to searches, such as images, tables, and grids. It can even go a step further, coding a custom simulation or creating a tool in real-time, depending on the user’s prompt.
Gemini 3 Pro is currently available in Google Search for subscribers and in the Gemini app for all users.
