Gemini 3 is the latest generation of multimodal artificial intelligence (AI) models developed by Google DeepMind. Announced on November 18, 2025, this model is the successor to Gemini 2.5 and is described by Google as their "most intelligent" model as well as "the world's best for multimodal understanding". This launch marked a significant step in Google's effort to lead the generative AI race, with the model integrated directly into various of their flagship products, such as Google Search and the Gemini app, on the same day of launch.
Core Capabilities
Reasoning and Multimodal Understanding
Gemini 3 is built with state-of-the-art reasoning capabilities designed to understand the depth, nuance, and context behind user requests. Unlike its predecessors, which often required explicit format instructions, Gemini 3 introduces a "generative interface". This feature allows the model to determine the best output format itself, such as creating an immersive visual layout, a diagram, or even a simple animation, if it is considered more effective than ordinary text. The model also shows high performance on various benchmarks, including a score of 1501 Elo on the LMArena leaderboard, and a score of 37.5% on the Humanity's Last Exam test that measures PhD-level reasoning.
Agentic Capabilities and "Vibe Coding"
One of the main improvements in Gemini 3 is its capability as an agentic AI. The model can not only respond to questions, but also plan and execute multi-step tasks independently. This capability is strengthened by the introduction of Google Antigravity, a new development platform that allows developers to delegate complex tasks, such as writing, testing, and verifying code, to AI agents that can work across the editor, terminal, and browser. This approach of enabling the creation of an entire interface or application from a single natural language prompt is known as "vibe coding".
Model Variants
Gemini 3 Pro
Gemini 3 Pro is the flagship variant designed to handle complex tasks, deep reasoning, and creative content generation. This model is available to developers through Google AI Studio and Vertex AI, as well as to Google Cloud customers. Pro shows significant improvements in visual understanding, code generation, and performance on long-running tasks. The model also serves as the foundation for various new features in Google Search and the Gemini app.
Gemini 3 Flash
Gemini 3 Flash is a faster and more efficient variant, designed for high-frequency tasks that require greater speed and lower cost. Although lighter, Flash still offers academic-level reasoning capabilities and excels at agentic coding tasks, with a score of 78% on the SWE-bench Verified benchmark. This model is the standard model in the Gemini app for everyday use.
Gemini 3 Deep Think
The Gemini 3 Deep Think feature is a special reasoning mode that pushes the model's capabilities further by exploring several hypotheses in parallel to solve highly complex problems, such as scientific research, advanced mathematics, or intricate planning. This mode was first made available to Google AI Ultra subscribers.
Impact and Reach
With the launch of Gemini 3, Google emphasized a "Google scale" approach by bringing the model simultaneously to various platforms. This includes integration in Google Search, the Gemini app, developer platforms such as AI Studio and Vertex AI, and the new Antigravity platform. The company also offered one year of free access to Gemini Pro for students in the United States to encourage adoption in academic circles. This step reflects Google's strategy of not only developing advanced technology, but also ensuring broad accessibility and integration in everyday life and the professional world.
References
· MIT Technology Review. Google’s new Gemini 3 “vibe-codes” responses and comes with its own agent.
· Google Cloud Blog. Gemini 3 is available for enterprise.
· Bloomberg. Google Launches New Gemini AI Model With Interactive Answers.
· Google Blog. Gemini 3 Membuka Era Kecerdasan Baru.
· CNN Indonesia. Google Rilis Model AI Gemini 3, Bisa Apa Saja?