Google unveils Gemini 3 AI model and AI-first IDE called Antigravity

Google is advancing its Gemini initiative with the introduction of the upgraded Gemini 3 model, which is now available in a limited capacity. This new version promises enhanced visual outputs and improved accuracy, addressing previous concerns about misinformation. The company claims that Gemini 3 sets a new standard for coding efficiency and creativity. Alongside Gemini 3, Google has unveiled an AI-first integrated development environment (IDE) named Antigravity, designed to facilitate coding using the latest advancements in artificial intelligence. This launch is viewed as another significant milestone on the journey toward achieving artificial general intelligence (AGI). Gemini 3 boasts expanded capabilities in simulated reasoning, offering a better grasp of text, images, and video content. Initial feedback from testers indicates that the model is performing exceptionally well, with Gemini 3 currently leading the Language Model Arena leaderboard, boasting an ELO score of 1,501, surpassing its predecessor, Gemini 2.5 Pro, by 50 points. While factual accuracy remains a challenge for generative AI models, Google asserts that Gemini 3 marks a significant improvement. In a recent evaluation involving 1,000 questions, the model achieved a record score of 72.1 percent on the SimpleQA Verified test. Although this indicates that it still misinterprets nearly 30 percent of general knowledge inquiries, the company sees this as a noteworthy advancement. In more rigorous assessments, such as Humanity’s Last Exam, which evaluates PhD-level knowledge, Gemini 3 set a new benchmark with a score of 37.5 percent, all without external tools. The model has also excelled in mathematics and coding, achieving remarkable scores in the MathArena Apex and WebDev Arena tests, reinforcing its capabilities in software development tasks.