Google has released a preview of Gemini 3.1 Pro, an upgraded version of its flagship artificial intelligence model, with a full public launch expected soon. The company said the new system represents a significant leap from Gemini 3, introduced in November 2025, and confirmed that the preview is already rolling out across consumer, developer and enterprise platforms.
In a statement, Google described Gemini 3.1 Pro as the core intelligence behind recent advances delivered under its Gemini 3 Deep Think update, which targeted complex challenges in science, research and engineering. The company said it is deploying the new model across everyday applications to bring stronger reasoning capabilities to users. Developers can access the preview through the Gemini API in Google AI Studio, Gemini CLI, Google Antigravity and Android Studio, while enterprise clients can integrate it via Vertex AI and Gemini Enterprise. Consumers are receiving the upgrade through the Gemini app and NotebookLM.
Google reported that Gemini 3.1 Pro scored 77.1 per cent on ARC AGI 2, a benchmark designed to test a model’s ability to solve entirely new logic patterns rather than rely on memorised examples. According to the company, the result more than doubles the reasoning performance recorded by Gemini 3 Pro. On another evaluation known as Humanity’s Last Exam, Google said the model outperformed Anthropic’s Claude Opus 4.6 and OpenAI’s GPT 5.2, underscoring intensifying competition among leading artificial intelligence labs.
Independent testing platform APEX Agents ranked Gemini 3.1 Pro first for professional task performance. Brendan Foody, chief executive of Mercor, which operates the APEX system, said on social media that the model now leads the leaderboard, noting the pace at which AI agents are advancing in real knowledge work. Google described the upgrade as a stronger foundation for complex problem solving, capable of handling layered instructions, multi step workflows and detailed data synthesis beyond basic text generation.
The release comes just three months after Gemini 3 and arrives amid rapid product cycles from rivals including OpenAI and Anthropic. Google DeepMind has published a model card outlining Gemini 3.1 Pro’s intended uses, limitations and safety measures. The company said the system is natively multimodal, meaning it can process text, images and other inputs, expanding its applications while raising fresh content moderation considerations. For now, the preview phase will allow further testing ahead of general availability, with higher usage limits for Google AI Pro and Ultra subscribers and NotebookLM access restricted to those plans.