Google has unveiled its next-generation AI model, Gemini 3. Gemini 3 Pro is considered the most intelligent model Google has ever developed, boasting significantly improved fact-based response capabilities. It will be available in the Gemini app and search AI mode from day one.
The Gemini 3 Pro is built on a multimodal architecture that simultaneously understands text, images, and audio to seamlessly process a single request. Its application ranges significantly, from everyday use to specialized fields, such as reading handwritten recipes into a cookbook or analyzing long video lectures to generate study flashcards. Search also offers a generative UI combined with responsive graphical simulations, enabling users to organize and present complex information at a glance.
Performance indicators have also significantly improved. It ranked first on the human-evaluation-based platform LMArena and surpassed existing models and competitors in several benchmarks measuring high-difficulty reasoning, demonstrating overall intelligence improvements. It also demonstrated strengths in math, science, and image comprehension, establishing the foundation for providing high-quality answers in a variety of fields.
The biggest change is the expansion of agent functionality. Gemini 3 has secured the stability to independently design and execute complex, multi-step workflows, including long-term planning. Google has begun experimenting with Gemini agents handling repetitive tasks like organizing Gmail, researching travel, and booking travel appointments.
There are also significant changes for developers. Along with Gemini 3, Google has launched Google AntiGravity, a new agent-centric development environment. This environment combines a chat terminal browser into a single space, allowing agents to continuously process code, execution, and validation, significantly streamlining the developer workflow.
Google emphasized that Gemini 3 has been tuned to reduce unnecessary sycophantic responses and deliver essential information concisely. Its security assessment has also been strengthened, improving its ability to respond to prompt injection attacks and cyber threats.
Gemini 3 Pro is available to everyone in the Gemini app starting today, and in the US, it's also available in Search AI mode for Google AI Pro and Ultra subscribers. Developers can leverage the model through AI Studio Vertex AI Gemini CLI Antigravity.
