Gemini 3.1 Pro Preview is an upgraded foundation model from Google, released in February 2026 as a refinement of the Gemini 3 series. It is designed for complex problem-solving and tasks that require advanced reasoning, data synthesis, and nuanced understanding. The model is natively multimodal, supporting inputs across text, images, audio, video, and PDF documents with a 1 million token context window and a 64,000-token output limit.
A significant advancement in this version is its performance on abstract logic and reasoning benchmarks. On the ARC-AGI-2 benchmark, which evaluates a model's ability to solve entirely new logic patterns, Gemini 3.1 Pro achieved a verified score of 77.1%, more than doubling the reasoning performance of the standard Gemini 3 Pro. This is achieved through an upgraded core intelligence that includes a native "Thinking" mode for deeper deliberation on difficult prompts.
Technical Capabilities
The model is heavily optimized for agentic workflows and software engineering. It introduces improved tool orchestration, precise tool usage, and reliable multi-step execution. For developers, a specialized endpoint called gemini-3.1-pro-preview-customtools is available, which is specifically tuned to prioritize custom tools and bash execution. It also introduces a thinking_level parameter, including a "MEDIUM" setting, to allow users to optimize the trade-off between reasoning depth, cost, and speed.
In addition to reasoning, Gemini 3.1 Pro excels at creative coding and data visualization. It can generate functional, website-ready animated SVGs directly from text and build immersive 3D experiences. Its training and architecture focus on improved token efficiency and factual consistency, making it better at long-horizon stability and complex planning in specialized domains like finance and research.