Sundar Pichai Unveils Gemini 3.1 Pro as Google Pushes AI Reasoning to New Heights

Sundar Pichai has announced the launch of Gemini 3.1 Pro, the latest upgrade to Google’s flagship artificial intelligence model — and one that marks a major step forward in reasoning capabilities.

The new model scored 77.1% on ARC-AGI-2, a benchmark designed to test abstract reasoning against entirely new logic patterns. That result represents more than double the performance of its predecessor, signaling a significant leap in how Google’s AI handles unfamiliar and complex problems.

With Gemini 3.1 Pro, Google is emphasizing intelligence refinement rather than simply adding features. Notably, this is the company’s first-ever “.1” version release — previous mid-cycle updates typically used a “.5” increment. The naming reflects what Google describes as a focused upgrade to reasoning power, rather than a broad expansion of tools.

At its core, Gemini 3.1 Pro integrates the advanced reasoning engine introduced in Gemini 3 Deep Think, now made widely available across platforms including the Gemini app, NotebookLM, and developer services such as Google AI Studio and Vertex AI. It is also accessible via tools like Antigravity, Gemini CLI, Gemini Enterprise, and Android Studio.

Google says the model is designed for scenarios where “a simple answer isn’t enough.” In practical terms, that includes synthesizing large datasets into coherent summaries, solving multi-step scientific and technical problems, and supporting advanced coding tasks.

One standout feature is its ability to generate animated SVG files directly from text prompts. Unlike traditional video or raster graphics, these outputs are created purely in code, making them scalable without loss of quality and significantly lighter in file size. For developers and designers, this means cleaner visuals and more efficient performance.

Beyond ARC-AGI-2, Gemini 3.1 Pro posted strong benchmark results across other technical domains. It achieved 94.3% on GPQA Diamond, which measures scientific knowledge, 80.6% on SWE-Bench Verified for coding performance, and 85.9% on BrowseComp for agentic search capabilities. These metrics suggest improvements not only in reasoning but also in real-world task execution.

The model is launching in preview mode as Google continues validating performance, particularly around complex agent-based workflows. Consumers can access Gemini 3.1 Pro through the Gemini app and NotebookLM, with higher usage limits available for Google AI Pro and Ultra subscribers. Developers can integrate it via the Gemini API across Google’s ecosystem.

The announcement underscores Google’s ongoing race to lead in advanced AI systems, particularly in areas that require structured thinking and multi-step reasoning. As the competition in generative AI intensifies, Gemini 3.1 Pro positions Google as doubling down on depth of intelligence rather than surface-level novelty.

General availability is expected in the coming weeks, but the early benchmark scores already indicate that Google’s next phase of AI development is focused on smarter, more capable reasoning engines built for increasingly complex challenges.