Google DeepMind releases Gemini 3.1 Pro with doubled reasoning scores

Google DeepMind released Gemini 3.1 Pro today, a mid-cycle reasoning upgrade that scores 77.1% on ARC-AGI-2 -- more than double the performance of its predecessor, Gemini 3 Pro, on the same benchmark. The release marks the first time Google has used a ".1" version increment rather than its previous ".5" convention, signaling a new cadence for incremental capability updates. The model is available immediately in Google AI Studio, the Gemini API, Vertex AI, the Gemini App, and NotebookLM at unchanged pricing.

Previously, readers waiting for a meaningful reasoning improvement from Google's flagship model had to wait for a full version release. Gemini 3.1 Pro changes that calculus by folding intelligence developed during Gemini 3 Deep Think research directly into the production model. For anyone using AI to help with complex planning -- writing multi-step project outlines, debugging layered code problems, or analyzing lengthy documents -- the 40-60% relative improvement on complex planning tasks over Gemini 2.5 Pro is the headline capability gain. Pricing remains $2 per million input tokens and $12 per million output tokens, meaning existing users get the upgrade with no additional cost.