Google Stages Historic AI Comeback: Gemini 3 Dominates Benchmarks, Unveils Agentic Coding Platform Anti-gravity

Google has reportedly orchestrated a monumental “Mongolian reversal” in the AI landscape with the release of Gemini 3, transforming its narrative from an AI laggard to a formidable leader. Following the ill-fated Bard launch in February 2023, which saw a 9% stock dip due to immediate hallucinations, Google’s recent Gemini 3 debut on November 21, 2025, propelled its stock by 6% to an all-time high. Gemini 3 Pro has rapidly emerged as a frontrunner for the year’s best AI model, demonstrating exceptional performance across various benchmarks. It swept LM Arena leaderboards and secured top positions on Weird ML, Humanity’s Last Exam, and Math Arena, among others. Confirming its practical prowess, AMP has reportedly replaced Claude with Gemini 3 Pro to power its coding agent, citing superior capabilities over Sonnet 4.5.

Beyond the benchmark-shattering Gemini 3, Google also unveiled “Anti-gravity,” a new agentic development platform that signifies a strategic expansion into AI-powered coding environments. This VS Code fork, led by the former co-founders of Windsurf who joined Google in a $2.4 billion deal in July, facilitates agentic coding with Gemini. The platform, which positions developers as “managers of agents,” notably features Windsurf’s original coding agent, Cascade, as part of its core functionality. Anti-gravity enters a competitive field alongside Y Combinator-backed Chad, known for its “brain rot device” integration, and Google’s own Firebase Studio which also leverages AI agents for development. This move solidifies Google’s multifaceted approach to integrating advanced AI models into the software development lifecycle.