Google Unveils Gemini 3 AI Model with Deeper Understanding and New Agentic Tools
Key Points
- Gemini 3 is described as Google’s most intelligent AI model, better at grasping user intent.
- Multimodal features include turning video lectures into flash cards and analyzing sports footage.
- Integrated into AI Mode in Search and AI Overviews for Pro and Ultra subscribers.
- Gemini 3 Pro reduces sycophancy and improves security against prompt‑injection attacks.
- Google launches Antigravity, an autonomous agentic platform that plans and executes software tasks.
- Early access to advanced agentic capabilities is limited to $250/month AI Ultra subscribers.
- The rollout aims to address AI challenges like hallucinations while competing with other leading AI firms.
Google announced Gemini 3, its most advanced AI model to date, highlighting improved ability to grasp user intent and richer multimodal features. The model can transform long video lectures into interactive flash cards and analyze sports footage for performance insights. Gemini 3 will appear in AI Mode in Search, AI Overviews for Pro and Ultra subscribers, and powers new agentic platform Antigravity, which can autonomously plan and execute software tasks. The company also noted enhancements in security against prompt‑injection attacks and reduced sycophancy. Gemini 3’s advanced capabilities are initially available to Google AI Ultra subscribers.
Gemini 3: A New Leap in Understanding User Requests
Google introduced Gemini 3 as its most intelligent AI model yet, emphasizing that it is "built to grasp depth and nuance" and is better at understanding the intent behind a user’s request. The company highlighted multimodal abilities such as converting a long video lecture into interactive flash cards and analyzing a person’s pickleball match to pinpoint improvement areas.
Availability Across Google Services
Gemini 3 will not be limited to a single app. It will be integrated into AI Mode in Search and, for Pro and Ultra subscribers, into AI Overviews. In AI Overviews, the model can generate interactive elements that enhance the user experience.
Addressing AI Weaknesses
DeepMind leaders Demis Hassabis and Koray Kavukcuoglu noted that Gemini 3 Pro is less sycophantic, a problem that has plagued earlier models. The new version also offers stronger protection against prompt‑injection attacks, where bad actors attempt to make an AI ignore its original instructions.
Introducing Antigravity: An Agentic Development Platform
Google also unveiled Antigravity, described as an active partner that autonomously plans and executes complex software tasks while validating its own code. Antigravity works with Gemini 2.5’s Computer Use model for browser control and with the image model Nano Banana.
Subscription Model and Early Access
Gemini 3’s more powerful agentic capabilities will initially be available to $250/month Google AI Ultra subscribers. This tier will enable the Gemini Agent to perform multi‑step workflows, such as planning a travel itinerary.
Competitive Landscape and Market Context
The release comes amid an AI race involving OpenAI, Anthropic, and xAI. Google aims to address persistent challenges like hallucinations and sycophancy while demonstrating that its models can act on a user’s behalf. The company’s broader AI efforts are part of a larger market where AI firms represent a significant portion of major indices.