Google unveils Gemini 3.1 Pro: what’s new

Google has officially introduced its new flagship model, Gemini 3.1, the company announced in a blog post.

In the ARC-AGI-2 benchmark, which measures performance on unfamiliar reasoning tasks, the model achieved a verified score of 77.1%. That is well above the average human result of around 60%.

The previous version, Gemini 3 Pro released in November, scored 31.1% on the same test. This marks more than a 2.5× improvement in abstract reasoning in just three months.

According to Google, the new model leads across most benchmarks, outperforming systems such as Sonnet 4.6, Opus 4.6, and GPT-5.2. It demonstrates strong results across a wide range of tasks.

In the PhD-level science benchmark GPQA Diamond, the model reached 94.3%, while in the multilingual MMMLU test it scored 92.6%. On agentic coding tasks (SWE-Bench Verified), the model posted 80.6%.

Gemini 3.1 Pro is currently available in preview via the API, Google AI Studio, Vertex AI, and Antigravity. Subscribers to the Pro and Ultra tiers can also access it in the Gemini app and NotebookLM.

Google describes 3.1 Pro as “core intelligence” and recommends it for tasks such as large-scale data synthesis, visualization, and creative workflows.