On December 11 (local time), OpenAI introduced its cutting-edge model lineup, GPT-5.2. This series comprises three distinct versions: Instant, Thinking, and Pro, each meticulously crafted to cater to the diverse and demanding needs of professional work environments. The series showcases remarkable enhancements across a spectrum of tasks, including spreadsheet formulation, code generation, in-depth long-text comprehension, intricate multi-step reasoning, and sophisticated visual analysis.

Among these versions, GPT-5.2 Thinking stands out by achieving performance levels on par with or surpassing those of human experts in a staggering 70.9% of the knowledge work tasks evaluated across 44 different professions. Notably, it operates at a blistering speed, over 11 times faster than human experts, while incurring less than 1% of the associated costs. When it comes to coding prowess, GPT-5.2 Thinking shines brightly, scoring an impressive 55.6% on the SWE-Bench Pro test and reaching an 80% accuracy rate on the SWE-bench Verified test. Furthermore, GPT-5.2 has also made leaps and bounds in the realms of mathematical reasoning, abstract thinking, and visual perception.