Gemini 3.1 Pro Brings Google’s Reasoning Preview to Apps and APIs
Google released Gemini 3.1 Pro in preview across consumer and developer products. It reports a 77.1% ARC-AGI-2 result; NotebookLM access at launch was limited to Pro and Ultra subscribers.
Google released Gemini 3.1 Pro in preview on February 19, extending the model across the Gemini API, Vertex AI, the Gemini app, and NotebookLM. The practical significance is distribution: the same model generation is being offered through consumer and developer products, although access conditions differ.
Google describes 3.1 Pro as a step forward in complex, multi-step reasoning. Its model card lists agentic work, advanced coding, long-context and multimodal understanding, and algorithmic development among intended uses.
Claims vs checks#
Google reports a 77.1% ARC-AGI-2 score, more than double Gemini 3 Pro’s result. This is an attributed benchmark result, not evidence that every task improves by the same factor or that the model leads every independent ranking. The model card supplies the evaluation context and links its methodology.
At launch, the Gemini app rollout offered higher limits to Google AI Pro and Ultra subscribers. NotebookLM access was exclusive to Pro and Ultra users. Developer and enterprise access was in preview. Google said it wanted to validate the changes and further develop agentic workflows before general availability.
Limits#
Preview availability does not establish feature parity across the app, API, and enterprise products. This article makes no historical Arena placement claim and does not independently reproduce Google’s benchmarks.
Correction (October 14, 2026): Clarified launch access restrictions and separated Google’s benchmark claim from unverified historical rankings.