Google Just Launched Gemini 3.1 Pro, and the Reasoning Scores Are Impressive

Google’s AI just got a serious upgrade. Gemini 3.1 Pro rolled out on Thursday, and the benchmark numbers tell a pretty exciting story.

The company calls it “a step forward in core reasoning.” But what does that mean in practice? Smarter problem-solving, better creative coding, and some genuinely cool new capabilities that go well beyond what the previous version could handle.

ARC-AGI-2 Benchmark Scores Jump Dramatically

The headline number is hard to ignore. Gemini 3.1 Pro scored 77.1% on the ARC-AGI-2 test, which measures how well an AI solves abstract reasoning puzzles.

Compare that to Gemini 3 Pro’s score of 31.1% on the same test. That’s more than double the performance in one generation. Plus, Google says improvements show up across benchmarks, not just this one test.

The ARC-AGI-2 is considered one of the harder evaluations in AI research. So a jump that significant is worth paying attention to.

What Gemini 3.1 Pro Can Actually Do

Better benchmarks are great on paper. But Google also shared real-world examples that show where those improvements show up in everyday use.

First, there’s code-based animation. Gemini 3.1 Pro can generate animated SVG files from a simple text prompt. These files scale without losing quality and drop straight into websites. No design software needed, no manual coding.

Then there’s creative coding. Google demonstrated the model building an entire portfolio website based on a character from Emily Brontë’s Wuthering Heights. The site treated the character as a landscape photographer. That kind of contextual, imaginative output requires serious reasoning ability, not just pattern matching.

Perhaps the most striking example is interactive design. The model created a 3D interactive starling murmuration, where users can control the movement of the flock. A dynamic soundscape responds to how the birds move. It’s a wild demo, and it shows what complex multi-system generation looks like when reasoning improves.

Where You Can Access Gemini 3.1 Pro

The rollout starts now, though access depends on your plan. Gemini app users on AI Pro or Ultra plans get the new model starting Thursday.

NotebookLM subscribers on those same plans will also have access right away. So if you use NotebookLM for research or document analysis, the upgrade arrives automatically.

Developers and enterprise users can reach Gemini 3.1 Pro through the Gemini API. That includes AI Studio, Gemini Enterprise, Antigravity, and Android Studio. So the model slots into existing workflows without requiring major changes.

Rapid-Fire AI Competition Keeps Heating Up

Google’s timing is no coincidence. Just days before this release, Anthropic launched Claude Sonnet 4.6, which can operate a computer at a human baseline level. The AI model landscape shifts almost weekly right now.

That pace of competition benefits users most. Each new release pushes rivals to move faster, share more benchmark data, and demonstrate real-world capability rather than abstract claims.

Gemini 3.1 Pro’s jump from 31.1% to 77.1% on abstract reasoning is one of the cleaner examples of genuine progress in recent memory. Whether that translates to meaningfully better experiences in daily use is something people will test quickly. But the early signals look strong.

If you’re on an AI Pro or Ultra plan, it’s worth firing up a complex prompt today just to see what changed. The animation and creative coding features alone seem worth exploring.