GPT-5.2 closes in on human-level intelligence with groundbreaking performance
The GPT-5.2 artificial intelligence achieved scores of 53–54% on the most challenging AI test, ARC-AGI-2, almost matching the average human score of 60%. This marks a significant advancement in the AI’s cognitive capabilities.
Additionally, GPT-5.2 flawlessly solved one of the toughest math exams, AIME 2025, for the first time and set a new record of 70–74% on the GDPval test, which assesses real-world human tasks and often reflects the skill level of experienced professionals.
These results highlight rapid AI progress and suggest promising prospects for its use in complex analytical and professional domains in the near future.
