OpenAI introduced GPT-5.2, the smartest AI model yet, delivering performance gains in typing, coding and reasoning tests. The launch comes just days after CEO Sam Altman internally announced a “code red,” a company-wide push to improve ChatGPT in the face of intense competition from rivals.
“We announced this code red to really signal to the company that we want to pool resources in one particular area, and this is a way to really define priorities,” said Fidji Simo, OpenAI’s general manager of applications, during a briefing with reporters on Thursday. “We have overall increased resources focused on ChatGPT.”
Simo denied that OpenAI was rushing the release of GPT-5.2 because of the red code, saying the company had been working on releasing it for months. However, she found the additional ChatGPT resources “helpful.”
While OpenAI models and products were recognized as best-in-class following the launch of ChatGPT in 2022, this is no longer a done deal. The startup now faces a number of noteworthy competitors, perhaps none more formidable than Google, whose recently launched Gemini 3 model has been well received by the tech industry. Over the past year, Google’s Gemini app has grown at an impressive rate and now has over 650 million monthly vigorous users compared to OpenAI’s 800 million weekly vigorous users. These pressures have forced OpenAI to scale back some of its most ambitious projects, including work to bring advertising to ChatGPT, and refocus on improving its core technologies and products.
OpenAI calls GPT-5.2 its best model for everyday professional apply. GPT-5.2 Thinking achieved the highest scores to date on PKBval, an OpenAI benchmark that compares the performance of AI models and specialists in 44 real-world professions. The company claims that the model outperformed specialists in more than 70 percent of tasks and completed them 11 times faster.
Max Schwarzer, OpenAI’s post-training manager, says GPT-5.2 should also provide significant improvements in ChatGPT hallucinations. The company claims that GPT-5.2 Thinking induced hallucinations 38 percent less than GPT-5.1 in benchmarks measuring answers to fact-based questions.