Google has announced its entry into a new era with the launch of “Gemini 3”, describing it as its “smartest and most factually accurate” model to date. This launch represents a major step toward artificial general intelligence (AGI) and an opportunity for Google to pull ahead in the AI race, delivering a powerful model with advanced capabilities in multimodal understanding, coding, and planning, while focusing on a visual and interactive user experience that goes beyond traditional text responses.
Article Contents:
- The dawn of the Gemini 3 era
- Deep Think mode: New frontiers of intelligence
- Learn anything: The power of multimodal understanding
- Build anything: A revolution in coding
- Plan anything: Intelligent agents
- User experience: Generative interfaces and direct responses
- Availability and the future
- Frequently asked questions
The dawn of the Gemini 3 era
Google is kicking off this era with the preview release of “Gemini 3 Pro”, making it available across a range of Google products so users can utilize it in their daily lives for learning, building, and planning. For the first time, Google is giving everyone access to its flagship model within the Gemini app from day one.
Demis Hassabis, CEO of Google DeepMind, and Koray Kavukcuoglu, Chief Technology Officer, stated that Gemini 3 is “the world’s best model for multimodal understanding and the most powerful agentic coding model to date, offering richer visualizations and deeper interaction, all built on a foundation of sophisticated reasoning”.
Deep Think mode: New frontiers of intelligence
Google is also introducing “Gemini 3 Deep Think”, an enhanced reasoning mode that pushes Gemini 3’s performance even further. This mode aims to achieve a quantum leap in reasoning and multimodal comprehension capabilities to help users solve the most complex problems.
In testing, Deep Think mode demonstrated unprecedented performance. It outperformed Gemini 3 Pro on difficult benchmarks such as GPQA Diamond, achieving 93.8%, and scored 45.1% on ARC-AGI-2, demonstrating its ability to tackle novel challenges with finesse. This mode is currently available to safety testers before rolling out to Google AI Ultra subscribers.
Learn anything: The power of multimodal understanding
Gemini was built from the ground up to seamlessly integrate information across multiple modalities, including text, images, video, audio, and code. Gemini 3 pushes the boundaries of multimodal reasoning to help users learn in entirely new ways, leveraging a 1-million-token context window.
For example, if you want to learn your family’s traditional cooking, Gemini 3 can decode and translate handwritten recipes in different languages and convert them into a shareable family cookbook. If you want to learn a new subject, you can provide academic papers or long video lectures, and it can generate interactive flashcards or visualizations to help you master the material. It can even analyze video clips of a sports match, identify areas for improvement, and build a training plan for you.
Build anything: A revolution in coding
Gemini 3 is the best coding model Google has ever built, significantly boosting developer productivity. It is exceptional at handling complex prompts and instructions to deliver richer, more interactive web interfaces.
Gemini 3 tops the WebDev Arena leaderboard with a remarkable Elo score of 1487. It also scores 54.2% on Terminal-Bench 2.0, which tests a model’s ability to operate a computer via the terminal. Furthermore, it significantly outperforms previous models on the SWE-bench Verified benchmark with 76.2%. Developers can now build with Gemini 3 in Google AI Studio, Vertex AI, and platforms such as GitHub and Replit.
Plan anything: Intelligent agents
Gemini 3’s ability to reliably plan ahead over longer horizons has improved. The model tops the Vending-Bench 2 leaderboard, which tests long-term planning by managing a simulated vending machine business. Gemini 3 Pro maintained consistent decision-making for a full simulated year, achieving higher returns without drifting from the task.
The model also powers the experimental Gemini Agent feature, which can perform tasks on your behalf within the Gemini app, such as reviewing and organizing emails, or searching for and booking travel.
User experience: Generative interfaces and direct responses
Within the Gemini app, users can experience “generative interfaces,” a tool that allows Gemini 3 to create magazine-like visual layouts with browsable images, or dynamic layouts customized to your request, rather than just text responses.
In AI Mode within Google Search, Gemini 3 Pro will provide visual elements such as tables, grids, and interactive simulations based on your query.
Google also stresses that Gemini 3 Pro is less prone to empty flattery or “sycophancy,” in a clear nod to issues faced by rival models such as ChatGPT. Google describes the model’s responses as “smart, concise, and direct, replacing clichés and pleasantries with real insights—telling you what you need to hear, not just what you want to hear.”
Availability and the future
Gemini 3 Pro is available within the Gemini app for everyone starting today. Google AI Pro and Ultra subscribers in the United States can experience Gemini 3 Pro inside AI Mode. This launch marks a pivotal moment in the evolution of artificial intelligence, placing Google at the forefront of innovation in this fast-paced field.
Frequently asked questions
What is Gemini 3?
Gemini 3 is Google’s newest and most powerful artificial intelligence model, designed for multimodal understanding (text, images, video, audio) with exceptional capabilities in coding and planning.
What is the difference between Gemini 3 Pro and Deep Think mode?
Gemini 3 Pro is the core model available to the public and developers. Deep Think mode is an enhanced version with higher logical reasoning capabilities for solving extremely complex problems, currently available only to safety testers.
How can I use Gemini 3?
It can be used via the Gemini app, within Google Search (initially for subscribers), and through developer platforms such as Google AI Studio and Vertex AI.
