Google’s Gemini 3.1 Pro: A Deep Dive into the Latest AI Model

Google continues to rapidly advance its artificial intelligence capabilities with the release of Gemini 3.1 Pro, the newest iteration in the Gemini 3 series. Unveiled on February 19, 2026, Gemini 3.1 Pro is positioned as Google’s most advanced model for complex tasks, boasting improved reasoning and problem-solving abilities. This release builds upon the foundation laid by Gemini 3 Pro and is now powering an upgraded Deep Think tool.

What is Gemini 3.1 Pro?

Gemini 3.1 Pro is a natively multimodal reasoning model designed to comprehend vast datasets and tackle challenging problems from diverse information sources, including text, audio, images, video, and code repositories [Google DeepMind]. It leverages the architecture and data foundations of Gemini 3 Pro, offering a 1 million token context window for inputs and generating outputs of up to 64,000 tokens [Google DeepMind].

Deep Think: Enhanced Reasoning Capabilities

The core intelligence behind the recent improvements to Google’s Deep Think tool is Gemini 3.1 Pro [Ars Technica]. Deep Think is a specialized reasoning mode within Gemini Apps, aimed at solving complex problems requiring advanced reasoning. It is currently available to Google AI Ultra subscribers and Google AI Ultra for Business license holders [Google Help].

Performance Benchmarks

Google highlights several benchmark achievements for Gemini 3.1 Pro:

• Humanity’s Last Exam: Achieved a score of 48.4%, surpassing Gemini 3 Pro’s 37.5% and OpenAI’s GPT-5.2’s 34.5% [Ars Technica].
• ARC-AGI-2: More than doubled Google’s previous score, reaching 77.1% [Ars Technica].
• Codeforces: Attained an Elo rating of 3455, demonstrating significant advancements in competitive programming [9to5Google].
• International Math Olympiad 2025: Reached gold-medal level performance [9to5Google].

While Gemini 3.1 Pro demonstrates strong performance, it is slightly behind competitors like Claude Opus 4.6 in Arena leaderboard scores for both text and code [Ars Technica].

Practical Applications

The upgraded Deep Think, powered by Gemini 3.1 Pro, is designed for practical applications in science, research, and engineering [9to5Google]. This includes interpreting complex data, modeling physical systems through code, and even converting sketches into 3D-printable designs [9to5Google].

Using Deep Think

To access Deep Think, users must be a Google AI Ultra subscriber or have a Google AI Ultra for Business license and be 18 years or older [Google Help]. Users can access it through gemini.google.com by entering a prompt in the text box and clicking “Deep Think Submit” [Google Help]. It’s important to note that Deep Think is an experimental feature and may be discontinued or suspended at any time [Google Help].