Elevating mathematical and algorithmic rigor
Specialized versions of Deep Think mode have successfully navigated some of the toughest challenges in reasoning, achieving gold-medal standards at math and programming world championships.
Gemini 3.1 Deep Think
Best for modern challenges across science, research and engineering
Our most specialized reasoning mode is now built on top of Gemini 3.1 Pro, our most intelligent model yet.
Pushing the boundaries of intelligence
By blending deep scientific knowledge with everyday engineering utility, Deep Think mode moves beyond abstract theory to drive practical applications
Navigating complex scientific domains
Deep Think mode excels on complex problem solving in broad scientific domains such as chemistry and physics.
Accelerating real-world engineering
Deep Think mode is built to drive practical applications, like helping accelerate rapid prototyping, evaluate experimental data, and optimize complex system designs.
Hands-on
Explore how Deep Think mode is already solving complex challenges across science and engineering
Slide 1 of 3
Optimizing 2D semiconductor fabrication
At Duke University, the Wang Lab utilized Deep Think mode to solve a complex challenge in materials science: optimizing fabrication methods for crystal growth.
Identifying logical errors in complex mathematics research
Lisa Carbone, a mathematician at Rutgers University, used Deep Think mode to review a specialized paper in the field of high-energy physics and infinite-dimensional algebra.
Accelerating mechanical engineering and rapid prototyping
Anupam Pathak, an R&D lead in Google’s Platforms and Devices division, tested Deep Think mode to accelerate the design and prototyping of complex physical components.
Performance
Reaching new heights across the most rigorous academic benchmarks
| Benchmark | Notes | Gemini 3.1 Deep Think Feb 2026 | Gemini 3 Pro Preview Thinking (High) | Opus 4.6 Thinking (Max) | GPT-5.2 Thinking (xhigh) |
|---|---|---|---|---|---|
| ARC-AGI-2 Abstract reasoning puzzles | ARC prize verified | 84.6% | 31.1% | 68.8% | 52.9% |
| Humanity's Last Exam Academic reasoning (full set, text + MM) | No tools | 48.4% | 37.5% | 40.0% | 34.5% |
| Search + code execution | 53.4% | 45.8% | 53.1% | 45.5% | |
| MMMU-Pro Multimodal understanding and reasoning | No tools | 81.5% | 81.0% | 73.9% | 79.5% |
| International Math Olympiad 2025 Mathematics | 81.5% | 14.3% | — | 71.4% | |
| Codeforces Coding and algorithms | No tools, Elo | 3455 | 2512 | 2352 | — |
| International Physics Olympiad 2025 (theory) Physics | 87.7% | 76.3% | 71.6% | 70.5% | |
| CMT-Benchmark Condensed matter theory | 50.5% | 39.5% | 17.1% | 41.0% | |
| International Chemistry Olympiad 2025 (theory) Chemistry | 82.8% | 69.6% | — | 72.0% |