Google Launches Gemini 3.1 Pro Model
Google has launched Gemini 3.1 Pro, its latest flagship AI model, which achieved a 77.1% score on the ARC-AGI-2 benchmark. The model is positioned for complex system synthesis and demonstrates advanced reasoning and agentic workflow capabilities. Early technical reviews highlight its improvements in tool use, planning, and multi-agent collaboration.
- The 77.1% score on the ARC-AGI-2 benchmark represents more than double the reasoning performance of its predecessor, Gemini 3 Pro. This benchmark is specifically designed to test an AI's ability to solve novel logic patterns it has never encountered, a key indicator of fluid intelligence. - For enterprise use, Gemini 3.1 Pro is being made available through Vertex AI and Gemini Enterprise. Developers can access the model via the Gemini API in Google AI Studio, as well as through tools like Antigravity, Gemini CLI, and Android Studio. - A novel feature is the model's ability to generate animated Scalable Vector Graphics (SVGs) directly from text prompts. Because these animations are code-based rather than pixel-based, they maintain perfect sharpness at any scale and have significantly smaller file sizes than traditional video formats. - The model is natively multimodal, capable of processing and reasoning across text, images, audio, and video inputs. It has a context window of up to one million tokens, allowing it to comprehend and work with large datasets and entire code repositories. - Google has specifically highlighted improvements in software engineering and agentic capabilities for domains like finance and spreadsheet applications. This is part of a broader focus on enhancing the model's ability to perform multi-step, autonomous tasks and utilize tools more effectively. - Compared to Gemini 1.5 Pro, the new model has more recent training data (January 2026 vs. August 2024) but is approximately 2.8 times more expensive for input and output tokens. - The release is part of a tiered rollout strategy, with Gemini 3.1 Pro initially available as a preview to gather feedback before general availability. This follows the release of Gemini 3 Deep Think in November 2025 and is part of a rapid series of updates to Google's flagship models. - In internal safety evaluations conducted under Google's Frontier Safety Framework, the model was tested in five key risk areas, including cybersecurity and harmful manipulation, and remained below critical risk thresholds in all of them.
