Google's Gemini 3 is live after months of hype. Here's what it can do. (www.businessinsider.com)

Google today launched Gemini 3, a major multimodal upgrade that Google says makes the model smarter, more creative and much better at combining text, images and video. DeepMind and Gemini product leads describe it as moving from “reading text and images” to “reading the room”: stronger reasoning, improved factuality (Google calls it its “most factual model” yet) and better coding abilities that let it generate presentations, interactive visuals and even full web-like interfaces from prompts. Google cites a 37.5% score on the 2,500-question “Humanity’s Last Exam” without tool use as evidence of improved math and science reliability. Crucially, Google is putting Gemini 3 into Search on launch day for paying US Pro/Ultra users via a new “Thinking” option and promises wider availability soon—showing Google’s distribution edge. The release also expands agentic features: an experimental “Gemini Agent” that can execute multi-step workflows across Google apps (Calendar, Gmail) and a new Antigravity platform for “vibe coding” and dynamic, generative interfaces. For the AI/ML community this signals progress in multimodal reasoning, tighter tool and app integration, and a push toward useful autonomous agents—advancements that change how models are evaluated (beyond benchmarks) and how they’ll be deployed in real-world products.