Gemini 3.1 Pro Preview

Gemini 3.1 Pro Preview is a Google reasoning model intended for complex software engineering and agentic workflows. A third-party model description highlights stronger software-engineering performance, improved agent reliability, and more efficient token use; the supplied evidence does not describe its architecture, training data, parameter count, or training scale. It is therefore best understood as a multimodal-capable preview focused on difficult reasoning rather than as an open-weight model.

The preview has credible benchmark signals in the supplied sources: Scalevise reports that Google highlighted a 77.1% ARC-AGI-2 score, while MarkTechPost describes a Gemini-SQL2 system powered by Gemini 3.1 Pro that reached 80.04% execution-verified accuracy on the BIRD single-model text-to-SQL leaderboard. These results indicate promise for tasks requiring reasoning, coding, schema-grounded data work, and tool-using agents, although the BIRD figure applies to a specific system rather than the model in every general-purpose use case.