Google unveils Gemini 3.1 Pro: the model surpasses human performance in ARC-AGI-2 test
Company. Google. announced the launch of a new artificial intelligence model Gemini 3.1 Pro, which, according to the company, demonstrated results above the human average in the ARC-AGI-2 test.
Test results
When passing an independent benchmark ARC-AGI-2, which assesses the ability of models to generalise and think logically, Gemini 3.1 Pro scored 77,1%.
For comparison, the average human score on this test is about 60%.
The company claims that the new version:
has surpassed previous generations of Gemini;
showed better results than competitive models, in particular GPT-5.2;
set record scores in science, programming and multilingual analysis tasks.
Where the model is available
Gemini 3.1 Pro is currently available in preview format via:
Gemini API;
Google AI Studio;
Vertex AI;
subscriptions Google AI Pro and Google AI Ultra.
Developers can integrate the model into their own products through Google's corporate and cloud services.
Context.
The race in the field of generative artificial intelligence between leading tech companies continues to heat up. Benchmarks such as ARC-AGI-2 are used to assess the level of abstract thinking of models and their ability to solve new problems without prior training on specific examples.
Experts point out that although test results are important, they do not always reflect the full picture of the practical effectiveness of AI in real-world conditions.
Ukrainian
English
