The Gemini 3 Deep Think model uses a reasoning system that evaluates and discards inconsistent logical paths before responding.
 |
| The Deep Think model successfully solved complex math, physics, and chemistry problems at an Olympic level. |
Google announced the release of Gemini 3 Deep Think, an enhanced version of its artificial intelligence model specializing in scientific reasoning and problem-solving. The update positions Deep Think as a key tool for researchers, engineers, and companies seeking advanced solutions in mathematics, physics, chemistry, and programming.
Unlike standard AI models, Gemini 3 Deep Think integrates a special reasoning technique that allows it to "think" before executing responses. When presented with a query, the model evaluates multiple logical paths and discards inconsistent ones before delivering a final solution.
This self-correcting mechanism, according to Google, makes a fundamental difference compared to other models, as it increases accuracy and reliability in complex tasks.
 |
| Researchers at Rutgers University used Gemini 3 Deep Think to detect logical errors in advanced mathematical papers. |
Deep Think was developed in collaboration with scientists and researchers from diverse fields, with the goal of creating an AI capable of tackling high-level problems. It excels in mathematics, advanced programming, and solving Olympic-level chemistry and physics challenges, as well as handling theoretical physics concepts and visual reasoning.
Superior performance in tests and practical applications
In initial benchmark tests, Gemini 3 Deep Think outperformed rival models such as GPT-5.2 and Claude Sonnet 4.5 in reasoning and scientific knowledge tasks. The model demonstrated greater accuracy in interpreting complex flowcharts and maintaining consistency across extensive contexts, even when analyzing documents hundreds of pages long.
During the demanding Humanity’s Last Exam assessment, which includes some of the most challenging problems in mathematics, science, and engineering, Deep Think performed exceptionally well, outperforming Claude Opus 4.6 by eight points and GPT-5.2 by an even larger margin. Furthermore, it achieved a score of 3455 in the Codeforces benchmark, demonstrating its ability to solve complex real-world programming tasks.
 |
| Gemini 3 Deep Think facilitates the analysis of complex data and the modeling of systems to support high-level scientific research. |
Gemini 3 Deep Think's scientific approach has already been tested in real-world settings. Lisa Carbone, a mathematician at Rutgers University, used the model to analyze a complex mathematical article. Deep Think identified a logical error that went unnoticed during peer review, demonstrating its usefulness as a tool to support the validation and improvement of scientific research.
In another case, the Wang Lab at Duke University used Deep Think to optimize processes in semiconductor manufacturing. The artificial intelligence designed a recipe for growing micrometric films, essential materials for the electronic devices of the future.
Availability and Access for the Scientific Community
Gemini 3 Deep Think is now available to AI Ultra tier users through the Gemini app. In addition, Google launched an early access program via the Gemini API, aimed at scientists, engineers, and companies interested in leveraging the model's advanced capabilities for their research and development projects.
Gemini 3 Deep Think is available in the Gemini app for AI Ultra users and via API for the scientific and business communities.
With the launch of Gemini 3 Deep Think, Google aims to solidify its leadership in developing artificial intelligence models geared towards science, engineering, and solving complex problems. The combination of advanced reasoning, precision, and the ability to interpret visual and textual information positions Deep Think as a fundamental tool for progress in various disciplines and technological innovation.
No comments:
Post a Comment