Gemini 4 Launches Worldwide

The Gemini 4 era has officially begun. Late tonight, Google CEO Sundar Pichai, Demis Hassabis and others almost simultaneously announced the worldwide release of Gemini 4 Argon, the new flagship model. Pro users can try it in the Gemini app and Search AI Mode, and developers can call it through paid APIs in AI Studio and Android Studio. Ultra users get Pro's features plus access in AI Studio and Antigravity. Gemini 4 Argon combines image, video and language understanding with reasoning and code execution. On 19 official benchmarks, the highest-thinking-level Argon ranked first alone on 13 and tied for first on one. Google also showcased Argon solving real engineering, science and robotics problems. Developer ecosystem head Logan Kilpatrick said the Gemini 4 team worked for six months, and over 100,000 Google employees have already tested it internally. Highlights include an AI working through the night for 11 hours to pass 16,892 unit tests and compress a months-long Chrome architecture refactor into weeks; Gemma 4 E2B inference running 2.85 times faster with consistent output tokens; and five AI-assisted novel enzymes all passing lab catalytic activity verification. Argon also wrote a ray tracer from scratch in Python and NumPy to render a sunlit library, turned a photo into an interactive mini racing track, and after learning from a human video, hit the target seven times in a row on a real Apollo robot. It offers a 1-million-token workspace, with launch pricing of $2 and $10 per million input and output tokens through November 30, reverting to $4 and $20 afterward. In fusion research, it helped researchers find a broader class of exact solutions challenging the 1967 Grad conjecture.