OpenAI has recently released an updated version of ChatGPT, GPT-5.2, calling the new AI model its most advanced for long-running agents and professional knowledge work.
According to the maker of ChatGPT, the new model of chatbot professionally excels at science, math, and coding benchmarks.
The release came on the heels of CEO Sam Altman’s decision to declare a “code red” at the company in an effort to overhaul and improve the quality of ChatGPT.
Given the intensity of competition faced by the tech giants, such as Google and Anthropic, OpenAI has decided to improve personalization features for users, enhance its reliability and speed, and expand its searching capacity for a wider range of questions.
As per an internal memo, Altman also instructed the staffers to delay other planned initiatives, including advertising, the integration of AI-powered shopping tools and personal assistant ChatGPT Pulse.
The decision to impose red code was prompted by Google’s newly released Gemini 3 AI model, which outperformed ChatGPT on certain benchmarks including logic puzzles, expert-level knowledge, image recognition, and math problems.
Moreover, Anthropic has also become popular among businesses and enterprises due to its remarkable coding capabilities.
As a result, Fidji Simo, OpenAI’s CEO of applications said “We designed 5.2 to unlock even more economic value for people. It’s better at creating spreadsheets, building presentations, writing code, perceiving images, understanding long contexts, using tools and then linking complex multi-step projects.”
GPT-5.2 represents a significant leap in AI capabilities, especially in professional, complex, and agentic tasks.
GPT-5.2 Thinking is the first model from OpenAI to perform at or above a human expert level on GDPval benchmark, measuring well-specified knowledge across 44 occupations.
The model also demonstrates significant gains in coding and agentic performance with 55.6 percent accuracy in multi-language and 80 percent in Python.
The AI model also shows substantial improvements in reasoning and factuality, characterized by fewer hallucinations than its predecessor. The model also excels at long-context reasoning, making deep analysis of research papers, reports, and long contracts.
GPT-5.2 accomplishes 100 percent on the AIME 2025 competition math benchmark, 93.2 percent on the graduate-level GPQA Diamond benchmark.
GPT-5.2 Thinking sets a new state of the art on FrontierMath (Tier 1–3), solving 40.3% of problems.
According to Arun Chandrasekaran, an analyst at market research and IT consulting firm Gartner, “GPT-5.2 shows improvements in reasoning, coding and working with a range of inputs from text to audio, video, and more—all areas where OpenAI has faced challenges from Google and Anthropic.”
As per Ray Wang’s observations, founder of Constellation Research, GPT-5.2 capabilities make it a potential rival to Gemini 3, but it still lacks in performance to reverse Google’s momentum.
“For businesses, what OpenAI did was make it easier to create office productivity tools, Gemini is still more integrated,” Wang said.
“Compared to the prior version of its model, GPT-5.1, GPT-5.2 represents a huge jump in a variety of professional-worker tasks,” said Aaron Levie, CEO and co-founder of Box.
