Google’s Gemini 3 Arrives to Challenge OpenAI’s GPT-5, Better Understands Everyday Questions

Google just released Gemini 3 and it’s no coincidence. OpenAI’s GPT-5 is still fresh in everyone’s minds and now Gemini 3 is available in the app for anyone to use and promises to cut through the noise and give you exactly what you want without back and forth.

Gemini 3 is finally putting together text, images, video and audio in a way that previous versions only hinted at. Want to upload that dodgy video of your pickleball game? It breaks down your swing form, flags any footwork issues and even sketches up a basic training drill for you, complete with hand drawn diagrams. Or hand it that crumpled up recipe scribbled on notebook paper and it scans the handwriting, deciphers the ingredients and turns out a digital cookbook version complete with step-by-step photos and all that good stuff. That’s due to a 1 million token context window, which allows the model to hold a large amount of data without losing its train of thought. On video understanding benchmarks, it achieves an accuracy rate of 87.6%, which is more than slightly higher than the competition, resulting in fewer errors when dealing with real-world difficulties such as overlapping conversation and quick cuts.

Developers can tell it something high-level like “Build a retro 3D spaceship game” and Gemini 3 will generate the entire script, all the assets, debugging comments and even suggestions for smoother gameplay. This “vibe coding” lets you explain what you’re after in plain English while the AI fills in all the technical details. It scored a respectable 76.2% on SWE-bench, the software engineering work test, and a meaty 1487 Elo on WebDev Arena. To top it all off, Google has linked this with Antigravity, a new platform that allows AI agents to manage whole processes from wireframe sketching to browser testing on their own. You describe the app, and it will plan, code, and validate it for you.

Gemini 3 excels at long-term orchestration, as when you ask it to sort your overflowing inbox, it creates responses, schedules follow-ups, and flags trends such as recurring vendor issues for bulk treatment. In planning mode, it manages multi-step chains such as booking a trip by comparing flights, lodgings, and local events while staying inside your budget. In planning mode, it manages multi-step chains such as booking a trip by comparing flights, lodgings, and local events while staying inside your budget. A benchmark called Vending-Bench explains how to run a simulated business for a year, making decisions that boost earnings while staying on schedule.

Gemini 3 Pro is now available in the app for everyone, while Pro and Ultra users can access it through Search’s AI Mode, which includes interactive components such as custom calculators and simulations. Developers gain access through APIs and technologies such as Vertex AI, while organizations receive unique processes. Top-tier plans will gain access to more advanced capabilities such as agentic planning.