Google has rolled out Gemini 3 Deep Think, a new AI model with advanced reasoning tasks in science, engineering, and research. The update is part of the Gemini 3 lineup, designed to work on problems with incomplete data and no clear solutions.
Gemini 3 Deep Think is available to Google AI Ultra subscribers through the Gemini app. It is also being offered via the Gemini API to select researchers, engineers, and enterprises under an early access program.
According to Google, the model showed strong results in global benchmarks. It scored 48.4 percent on Humanity’s Last Exam, 84.6 percent on ARC-AGI-2, and reached an Elo rating of 3455 on Codeforces, a competitive programming platform. It also delivered gold-medal level performance at the 2025 International Math Olympiad and excelled in physics and chemistry Olympiads.
Also Read: Google Gemini can now help you code your first app
Beyond benchmarks, Gemini 3 Deep Think is built for practical use. Researchers can use it to interpret complex data, while engineers can model physical systems. Early testers have used the model to spot subtle flaws in advanced mathematics papers.
What do you think of Google’s Gemini 3 Deep Think? Will it speed up research? Stay tuned.
