Google releases Gemini 3.1 Pro, doubles reasoning, scores 77.1% ARC-AGI-2
Technology
Google has released Gemini 3.1 Pro, its latest AI model that is way better at solving tricky problems.
It scored 77.1% on the ARC-AGI-2 benchmark, basically doubling its reasoning power compared with the previous version.
This follows its recent "Deep Think" update, which focused on tackling tough scientific questions.
Gemini 3.1 Pro HLE 44.4% available
Gemini 3.1 Pro also outperformed rivals on Humanity's Last Exam (HLE) with a 44.4% score, though Anthropic's Claude Opus 4.6 still leads in text skills, according to the Center for AI Safety.
Developers can try Gemini 3.1 Pro early via Google AI Studio or Android Studio, while regular users get access through NotebookLM and the Gemini app, making this upgrade easier to explore for everyone interested in next-generation AI tools.