Google suddenly released Gemini 3.1 Pro: core inference performance has doubled directly

Google has just unveiled its latest foundational model, Gemini 3.1 Pro.

Last week, Gemini 3 Deep Think, aimed at research engineering scenarios, was just released.

Today’s release is the underlying intelligence engine that supports its technological breakthroughs.

Compared to the Gemini 3 Pro released last November, the new model's inference performance has doubled.

In the ARC-AGI-2 evaluation, which assesses the model's ability to crack new logical paradigms, Gemini 3.1 Pro scored 77.1% (while 3 Pro scored 31.1%). Most inference tasks are state-of-the-art (SOTA), and its coding capability is very close to Opus 4.6, with SWE-Bench validating at 80.6%, very close to Opus 4.6's 80.8%. It looks strong, but these numbers are just for reference; actual usage will truly test its capabilities, and I believe everyone will reach conclusions soon.

It focuses on multi-source data integration, complex visual analysis, and high-difficulty task decomposition.

Starting today, the new model will be fully pushed out in preview version.

Consumer users can directly access it through the Gemini application and NotebookLM.

Google AI Pro and Ultra subscribers have higher usage limits.

They can directly pull the Gemini API in Google AI Studio.

The supporting toolchain, Gemini CLI, and Android Studio have completed the first batch of adaptations.

Google Antigravity is also online.

Enterprise customers can access testing through Vertex AI and Gemini Enterprise.

Risk Warning and Disclaimer

The market has risks, and investments should be made cautiously. This article does not constitute personal investment advice and does not take into account individual users' specific investment goals, financial situations, or needs. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investing based on this is at your own risk