Overview
- Gemini 3 powers Google’s AI Mode in Search from day one and is live in the Gemini app, with Pro features in-app and in Search for AI Pro and Ultra subscribers.
- Developers get access via APIs in AI Studio, the Gemini CLI and Vertex AI, alongside Antigravity, an IDE for building and supervising multi‑step agents that includes Computer Use and the Nano Banana image tool.
- Google and DeepMind cite top results on LM Arena and improvements on Humane’s Last Exam, with advances in coding performance and logical reasoning.
- The model extends multimodal capabilities across text, images, video, audio and code, enabling interactive layouts, simulations and graphics in responses.
- Google says this is its most thoroughly tested model to date with stronger defenses against prompt injection and misuse, as Sundar Pichai also warned of potential overheating in the AI investment cycle.