OpenAI Group PBC has officially launched its latest large language model, GPT-5.2, marking a significant advancement in artificial intelligence capabilities. Released today, the model is available in three variations: Instant, Thinking, and Pro. The company claims that the Thinking and Pro versions deliver unprecedented performance, particularly in mathematical tasks, surpassing competitors in various domains.
In testing the mid-range Thinking version using the FrontierMath benchmark, which includes challenging college-level math problems, GPT-5.2 Thinking achieved a record success rate of 40.3% in solving these problems correctly, a new industry benchmark. Notably, the model secured a perfect score on a qualifying examination for the International Mathematical Olympiad.
The Pro version of GPT-5.2 demonstrated its prowess by assisting researchers in a groundbreaking discovery within the field of statistical learning theory. It effectively resolved a simplified form of an open problem that had been presented during a 2019 math conference, accomplishing this without human guidance on how to approach the task, according to OpenAI.
Compared with its predecessor, GPT-5.1, the new model exhibits enhanced capabilities in interpreting charts commonly found in scientific literature. Testing using the CharXiv Reasoning benchmark revealed that the Thinking version correctly interpreted 88.7% of charts, representing an improvement of over 8% from GPT-5.1 Thinking.
The visual reasoning functionalities of GPT-5.2 extend to various applications. In internal evaluations, OpenAI staff utilized the model to identify key components from a low-resolution image of a motherboard. Additionally, it has shown competency in analyzing business intelligence dashboards and product diagrams.
Front-end development tasks, including the creation of visual application components such as interfaces, also benefit from GPT-5.2’s enhanced capabilities. The model excels in developing three-dimensional assets, including simulations. In other programming tasks, it achieved an impressive score of 55.6% on the SWE-Bench Pro, a benchmark encompassing complex coding challenges across multiple programming languages, and scored 80% on the Python-specific SWE-bench Verified version.
OpenAI began rolling out GPT-5.2 to ChatGPT users today and has made the model accessible via its application programming interface for developers. For pricing, the entry-level model is set at $1.75 per million input tokens and $14 per million output tokens. The Pro version commands higher fees, priced at $21 and $168 per million input and output tokens, respectively. Notably, OpenAI indicated that developers could slash output costs by up to 90% using a caching feature that retains frequently requested answers, minimizing the need for repeated generation.
The introduction of GPT-5.2 positions OpenAI at the forefront of advancements in artificial intelligence, particularly in mathematical reasoning and programming tasks. As the company continues to innovate and expand the capabilities of its language models, the implications for various industries and applications are significant, potentially reshaping how businesses and researchers approach problem-solving in complex fields.
See also
Rivian Unveils Advanced AI Strategy with Custom Processor, Plans LiDAR for 2026 R2 Models
Elon Musk’s xAI Launches Grok Chatbot in 5,000 El Salvador Schools for 1M Students
BCC Research Reveals $1.5T AI Market Surge Driven by Biotech, Food & Beverage Innovations
GenAI’s Limits in Scientific Discovery: New Study Reveals Creative Shortcomings
US Customs Proposes Mandatory 5-Year Social Media History for 42 Countries’ Tourists
