Google's Gemini 3 Deep Think achieves gold medal standards in Math and Physics Olympiads

Google Says Deep Think AI Can Partner on Advanced Math, Science

Alphabet Inc. has updated its Gemini Deep Think artificial intelligence model for better performance in math and science research, the company said. After close partnership with researchers, the specialized reasoning model is now able to help scientists move from theoretical reasoning to practical applications, according to a blog post. The AI model uses Google's search to avoid inaccuracies and wrongful citations when doing research, according to a separate blog post. Gemini 3 Deep Think, as the model is called, can also help researchers in chemistry, computer science and physics. The new model is part of a push by the leading AI developers to build more advanced tools that can field everything from complex coding to scientific research. Anthropic, for example, recently released a new version of its most powerful AI model to do both financial research and legal services, leading to a market selloff for traditional software firms. Google built out a math research agent, dubbed Aletheia, that can conduct autonomous research or collaborate with humans. The new agent can also "admit failure to solve a problem," which improved efficiency for researchers, Alphabet's Google said. Google published some of the papers that resulted from the new technology. "Spanning diverse fields -- from information and complexity theory to cryptography and mechanism design -- the results demonstrate how AI is fundamentally shifting research," the company said. Google said it is available in the Gemini app for Google AI Ultra subscribers and also for select researchers, the post said.

Gemini 3 Deep Think gets 'major upgrade' aimed at practical applications

Deep Think is Gemini's "specialized reasoning mode," and Google today announced a "major upgrade" to let it "solve modern challenges across science, research, and engineering." Google worked with scientists and researchers on this update, with the goal of using Deep Think to "tackle tough research challenges" that "often lack clear guardrails or a single correct solution and data is often messy or incomplete." By blending deep scientific knowledge with everyday engineering utility, Deep Think moves beyond abstract theory to drive practical applications. In terms of benchmarks for this Gemini 3 Deep Think upgrade, Google highlights: This leap in mathematics and competitive coding is joined by boosted performance in chemistry, physics (including theoretical), and other scientific domains. Practical applications for Deep Think allow "researchers to interpret complex data, and engineers to model physical systems through code." With the updated Deep Think, you can turn a sketch into a 3D-printable reality. Deep Think analyzes the drawing, models the complex shape and generates a file to create the physical object with 3D printing. This Gemini 3 Deep Think upgrade is now available in the Gemini app for Google AI Ultra subscribers, while Google is also making it available via the Gemini API (express interest for early access here) for enterprise users.

Google Gemini 3 Deep Think hits gold medal standards in math and physics olympiads

Google has officially unveiled a major upgrade to Gemini 3 Deep Think, its most sophisticated reasoning model designed to push the boundaries of intelligence in science, research, and engineering. This release marks a transition from general-purpose AI toward a specialized tool capable of navigating the nuances of advanced academia. While standard models often struggle with "messy" data or problems that lack a single clear-cut solution, Deep Think is built to thrive in these gray areas. By blending deep scientific knowledge with algorithmic rigor, Google is positioning this model as a critical collaborator for the global scientific community. Also read: Seedance 2.0: This Chinese AI video tool is outpacing Veo 3 and Sora 2 The most striking achievement of this updated model is its unprecedented performance on the world's most difficult academic benchmarks. Gemini 3 Deep Think has achieved gold-medal level performance in the 2025 International Math Olympiad, proving it can handle the abstract logic and creative problem-solving required at the highest levels of competitive mathematics. This expertise is not limited to numbers alone; the model also demonstrated gold-medal level results on the written sections of the 2025 International Physics and Chemistry Olympiads. These results suggest that the model has moved beyond mere pattern matching and is now capable of deep, first-principles reasoning. Beyond the classroom, the model is setting new industry standards for artificial general intelligence. It recorded a staggering 84.6% on the ARC-AGI-2 benchmark, a test specifically designed to measure fluid intelligence and the ability to learn new concepts on the fly. In the realm of competitive programming, it attained an Elo rating of 3455 on Codeforces, placing it among the elite tier of human coders. Perhaps most impressively, it scored 48.4% on "Humanity's Last Exam," a benchmark composed of questions specifically designed by experts to be nearly impossible for contemporary AI to solve without specialized tools. Also read: Microsoft warning: AI being brainwashed to favour some brands The true value of Gemini 3 Deep Think is already being realized in real-world research environments where human peer review often reaches its limits. At Rutgers University, a team used the model to review a highly technical mathematics paper focusing on the intersection of Einstein's theory of gravity and quantum mechanics. In a field where training data is scarce and the logic is incredibly dense, Deep Think successfully identified a subtle logical flaw that had remained unnoticed during traditional human peer review. This ability to act as a high-level auditor for scientific literature could fundamentally change how academic research is verified and published. Further practical success was seen at Duke University's Wang Lab, where researchers utilized the model to optimize fabrication methods for semiconductor materials. The model successfully designed a precise recipe for growing thin films larger than 100 micrometers, a target that had previously eluded researchers using standard methodologies. By modeling physical systems through code and interpreting complex datasets, Deep Think is proving that its reasoning capabilities have tangible benefits for material science and industrial engineering. Google is also showcasing the model's ability to bridge the gap between abstract design and physical manufacturing. One of the most practical new features allows Deep Think to analyze a simple hand-drawn sketch and transform it into a 3D-printable object. By understanding the geometry and physical requirements of the drawing, the model generates the necessary code to create a functional file for 3D printing. This capability streamlines the prototyping process for engineers, allowing for rapid iteration from a basic concept to a physical part. The updated Gemini 3 Deep Think is now available to Google AI Ultra subscribers within the Gemini app. To ensure this technology reaches the hands of those who can use it most effectively, Google is also launching an early access program for the Gemini API. This allows enterprises, researchers, and independent engineers to integrate these deep reasoning capabilities into their own custom applications. As AI continues to evolve, Google's latest offering suggests that the next great frontier isn't just about faster answers, but about more profound, verified logic that can solve the world's most complex scientific mysteries.