Under the leadership of Sundar Pichai, Google has unveiled Gemini 3 Deep Think, an upgraded AI model designed to advance complex reasoning across mathematics, coding, physics, and chemistry.
The model scored 48.4% on Humanity’s Last Exam, 84.6% on ARC-AGI-2, and achieved a 3455 Elo rating on Codeforces, signalling strong performance in abstract reasoning and competitive programming benchmarks. Gemini 3 Deep Think is now rolling out via the Gemini app and Gemini API, targeting researchers and enterprise users.
Strategically, this marks Google’s intent to move AI beyond conversational productivity into scientific discovery and high-order problem solving. As foundation models compete on reasoning depth rather than scale alone, benchmark credibility becomes a powerful signal of capability.
Overall, Google is positioning Gemini not just as an assistant—but as a research-grade reasoning engine for the AI era.
