Google released Gemini 3.1 Pro on Thursday, touting the update as a breakthrough in AI reasoning for tackling unfamiliar challenges. The model aced the ARC-AGI-2 benchmark with a verified 77.1% score, testing an AI’s ability to crack entirely new logic patterns without relying on trained examples.

Engineers at Google tuned this version specifically for multi-step problems where basic responses fall short. Officials described it as a ‘step forward in core reasoning,’ distinct from earlier Gemini 3 models. That benchmark leap signals the AI can now dissect complex topics, deliver visual breakdowns and synthesize scattered data into unified insights.

Developers gain immediate access through the Gemini API in Google AI Studio, the Gemini CLI, Android Studio and the Antigravity platform. Enterprise users tap into it via Vertex AI and Gemini Enterprise, ideal for reasoning across massive, diverse datasets in business settings.

Everyday users see the model roll out in the Gemini app and NotebookLM. Access links to AI Pro and Ultra subscription tiers in select markets, with boosted usage limits. Google calls this a preview phase to test refinements and collect user input before full deployment.

The push emphasizes reliability on tough tasks over casual conversation. Gemini 3.1 Pro supports advanced workflows, from research demanding deep analysis to creative projects needing structured ambition. Google positions it as a fresh standard for reasoning-intensive work across industries.

Early tests show the model excels at generating clear explanations for complex subjects. It pulls threads from varied sources into coherent narratives, aiding professionals in fields like data analysis and project planning. Subscription holders in supported regions already experiment with its enhanced limits.

Google’s broader strategy shines through this launch. The company integrates the model across consumer apps, developer tools and enterprise platforms. Feedback from the preview will shape its path to general availability, potentially redefining AI tools for high-stakes applications.