As OpenAI celebrates its 10th anniversary, the newly released GPT-5.2 series model has sparked widespread discussion. Official data shows that GPT-5.2 performs exceptionally well in multiple professional benchmark tests, even surpassing human experts in some areas, making it the AI model with the best performance in professional tasks to date.

According to OpenAI's introduction, GPT-5.2 has achieved technical breakthroughs in multiple fields. For example, in the GDPval test, the model scored 70.9% on 44 occupational tasks, exceeding top experts. Meanwhile, in the SWE-bench Pro programming test, GPT-5.2 achieved a SOTA (State of the Art) score of 55.6%, and the hallucination rate was reduced by 38% compared to the previous version, GPT-5.1. These achievements are encouraging, seemingly marking another leap in AI technology.

However, not all feedback has been positive. In the SimpleBench common sense reasoning test, GPT-5.2 scored lower than Anthropic's Claude Sonnet 3.7, especially performing poorly on some seemingly simple questions. For instance, the model often made mistakes when answering questions like "How many 'r's are in 'garlic'?" During three tests, it only got the answer right once. In contrast, competitors like Google's Gemini 3.0 could consistently pass these logical reasoning challenges. This has disappointed some users, and former AWS manager Bindu Reddy even stated directly: "It's not worth upgrading from GPT-5.1."

Although technological progress is undeniable, the challenges faced by GPT-5.2 also provoke reflection. The shortcomings of AI models in handling simple common sense issues have sparked debates about the level of AI intelligence. Does this mean a regression in certain aspects of technology, or is it just a normal phenomenon in the development process? In the future, OpenAI needs to further optimize and improve the model's performance in logical reasoning and common sense understanding.

The release of GPT-5.2 marks a significant advancement for OpenAI in professional fields, but it also reveals the model's shortcomings in basic tasks such as common sense reasoning. This debate about AI intelligence may become an important topic in future technological development.