Overview
- Google reports a 77.1% ARC-AGI-2 score for the model, more than doubling Gemini 3 Pro’s measured reasoning performance.
- Access begins today in the Gemini app and NotebookLM for Pro and Ultra users, with developer and enterprise preview in AI Studio, Antigravity, Vertex AI, Gemini Enterprise, the Gemini CLI, and Android Studio.
- The model supports a 1 million‑token input window and roughly 64,000–65,000 output tokens to handle large codebases, documents, and long-form generation.
- Gemini 3.1 Pro integrates the upgraded core intelligence from last week’s Deep Think update and enters preview to validate agentic workflows before general availability.
- Google highlights leading results on several reasoning and science tests, though independent leaderboards still place rivals like Anthropic’s Claude Opus 4.6 ahead on some text quality and safety measures; preview pricing remains unchanged at $2/$12 per million input/output tokens.