Google Gemini 3.1 Pro Unveiled: Advanced Reasoning and Code-Based Animation
Originally published in Turkish. Read the original →
Google has announced Gemini 3.1 Pro, its next-generation model designed to solve complex scientific and engineering problems. We take a closer look at this new release, with its advanced reasoning and its ability to generate SVG animations from text prompts.
Shifting up a gear in the AI race, Google has officially announced Gemini 3.1 Pro, the engine behind the Deep Think update that made big waves in the science and engineering world recently. Standing out especially for its ability to tackle complex problems and produce solutions, the model has been opened up to a broad audience, from developers to end users.
A Leap in Advanced Analysis and Reasoning
The biggest strength behind the new model is its advanced reasoning capacity. Built on the renewed foundation of the Gemini 3 series, 3.1 Pro posts a verified score of 77.1% on the demanding “ARC-AGI-2” benchmark. Compared with the previous version, Gemini 3 Pro, this result shows that its reasoning performance has more than doubled.
And it doesn't stop there; with a score of 44.4% on Humanity's Last Exam, a domain-based knowledge test considered one of the toughest exams in the tech world, Gemini 3.1 Pro manages to outperform many strong competitors on the market. So what do these scores mean in real life?
According to the company's statements, this new-generation AI doesn't just list correct answers. It can analyze huge datasets to extract holistic meaning, visualize abstract concepts to make them easier to understand, and offer creative solutions in projects that push the limits.
Performance Comparison Table
| Benchmark | Details | Gemini 3.1 Pro | Gemini 3 Pro | Sonnet 4.6 | Opus 4.6 | GPT-5.2 | GPT-5.3-Codex |
|---|---|---|---|---|---|---|---|
| Humanity's Last Exam (Academic reasoning) | No tools | 44.4% | 37.5% | 33.2% | 40.0% | 34.5% | — |
| Humanity's Last Exam (Academic reasoning) | Search+Code | 51.4% | 45.8% | 49.0% | 53.1% | 45.5% | — |
| ARC-AGI-2 (Abstract reasoning puzzles) | ARC Prize Verified | 77.1% | 31.1% | 58.3% | 68.8% | 52.9% | — |
| GPQA Diamond (Scientific knowledge) | No tools | 94.3% | 91.9% | 89.9% | 91.3% | 92.4% | — |
| Terminal-Bench 2.0 (Agentic terminal coding) | Terminus-2 | 68.5% | 56.9% | 59.1% | 65.4% | 54.0% | 64.7% |
| Terminal-Bench 2.0 (Agentic terminal coding) | Other best | — | — | — | — | 62.2% | 77.3% |
| SWE-Bench Verified (Agentic coding) | Single attempt | 80.6% | 76.2% | 79.6% | 80.8% | 80.0% | — |
| SWE-Bench Pro (Comprehensive agentic tasks) | Single attempt | 54.2% | 43.3% | — | — | 55.6% | 56.8% |
| LiveCodeBench Pro (Competitive coding) | Elo | 2887 | 2439 | — | — | 2393 | — |
| SciCode (Scientific research coding) | - | 59% | 56% | 47% | 52% | 52% | — |
| APEX-Agents (Long-horizon tasks) | - | 33.5% | 18.4% | — | 29.8% | 23.0% | — |
| GDPval-AA (Expert-level tasks) | Elo | 1317 | 1195 | 1633 | 1606 | 1462 | — |
| t2-bench (Agentic tool use) | Retail | 90.8% | 85.3% | 91.7% | 91.9% | 82.0% | — |
| t2-bench (Agentic tool use) | Telecom | 99.3% | 98.0% | 97.9% | 99.3% | 98.7% | — |
| MCP Atlas (Multi-step workflows) | - | 69.2% | 54.1% | 61.3% | 59.5% | 60.6% | — |
| BrowseComp (Agentic search) | Srch. + Py + Browse | 85.9% | 59.2% | 74.7% | 84.0% | 65.8% | — |
| MMMU Pro (Multimodal reasoning) | No tools | 80.5% | 81.0% | 74.5% | 73.9% | 79.5% | — |
| MMMLU (Multilingual Q&A) | - | 92.6% | 91.8% | 89.3% | 91.1% | 89.6% | — |
| MRCR v2 (8-needle) (Long-context performance) | 128k | 84.9% | 77.0% | 84.9% | 84.0% | 83.8% | — |
| MRCR v2 (8-needle) (Long-context performance) | 1M | 26.3% | 26.3% | Not supported | Not supported | Not supported | — |
Good News for Designers: Animations Generated with Code
One of the most innovative aspects of Gemini 3.1 Pro is the unique SVG generation capability it offers developers and content creators. The new model can turn simple text prompts from the user into lightweight, animated vector (SVG) files that can be embedded directly into websites.
Unlike the pixel-based video generation that has become widespread recently, these visual animations, produced entirely on pure coding principles, promise serious aesthetic quality and, thanks to their small file size, don't hurt web performance.
Where Can You Try the Model?
If you want to test these new features right away, the rollout has already begun:
- For developers: It can currently be tested as a preview release on the Google AI Studio and Antigravity IDE platforms.
- Enterprise users: For large-scale use, it can be integrated directly via the Vertex AI platform and Gemini Enterprise.
- Individual users: It is available in the standard Gemini app. However, for those who want to use the model's capacity to the fullest (Google AI Pro and Ultra subscribers), usage limits are set much higher. In addition, within NotebookLM only Pro and Ultra plan holders can test this power.
With its high-level reasoning algorithms, Gemini 3.1 Pro doesn't just stretch today's AI limits; it also lays a solid foundation for the more complex automation processes we will see in the days ahead.
