Google doubles the reasoning power of its core AI model with Gemini 3.1 Pro

Just one week after Google introduced the hyper-specialized “Deep Think” mode for scientific research, the team is back with a major upgrade to the engine that powers it all. Gemini 3.1 Pro is rolling out today, bringing the breakthroughs in core reasoning found in the Deep Think models to everyday consumer and developer applications and represents a significant leap in how the model handles logic and complex problem-solving.

On the ARC-AGI-2 benchmark—a rigorous test designed to see if an AI can solve entirely new logic patterns it hasn’t seen in its training data—Gemini 3.1 Pro achieved a verified score of 77.1%. To put that into perspective, that is more than double the reasoning performance of the previous 3 Pro model. For users, this means the AI is far more capable of “understanding” the intent behind a complex prompt rather than just predicting the next most likely words.

Intelligence applied to complex workflows

Google is positioning 3.1 Pro as the go-to model for tasks where a simple, one-off answer isn’t enough. One of the more compelling examples shared today involves Gemini 3.1 coding “a complex 3D starling murmuration. It doesn’t just generate the visual code; it builds an immersive experience where users can manipulate the flock with hand-tracking and listen to a generative score that shifts based on the birds’ movement.”

Availability and rollout

For those of us in the Google ecosystem, the rollout of 3.1 Pro is happening across a broad front. It is available starting today in the Gemini app and is also arriving as the exclusive engine for NotebookLM. For the developer crowd, the model is available in preview via the Gemini API, Android Studio, and a new agentic development platform called Google Antigravity.

If you are a subscriber to Google AI Pro or Ultra plans, you’ll see 3.1 Pro appearing with higher usage limits starting today. This update continues the rapid-fire release cycle we’ve seen from the Gemini team since November, and it’s clear that Google is focused on moving past simple chatbots toward truly agentic workflows that can actually perform complex, multi-step tasks with minimal hand-holding.