As was expected, on Thursday, OpenAI pushed out the latest update to its flagship AI model after CEO Sam Altman declared a “code red” within the company. The new release, GPT-5.2, is described by Altman as “the smartest generally-available model in the world, and in particular is good at doing real-world knowledge work tasks.” The update will start rolling out to paid users today and is already available in the API for developers.
According to OpenAI, GPT-5.2 sets a number of new highs across several benchmarks and supposedly “outperforms industry professionals at well-specified knowledge work tasks spanning 44 occupations.” In its “Thinking” mode, GPT-5.2 reportedly performs at or above human expert level on a number of tasks in which it is required to produce deliverables like blueprints, spreadsheets, and legal briefs. The company also claims that the update has reduced the amount of hallucinations that are produced by ChatGPT, and it allegedly produces 30% fewer response errors than its predecessor.
As for how GPT-5.2 compares to other models, it appears that it’s back in the mix for the top spot across a number of benchmarks, significantly surpassing Google’s Gemini 3 on the software development benchmark SWE-Bench Pro. That said, Gemini 3 still holds the top spot on much of the leaderboards on LMArena, a widely cited benchmarking tool used to compare LLMs. Google made waves last month with the release of its latest model and its notable leap in performance, not just compared to previous models but compared to competitors.
OpenAI has been less interested in drawing comparisons between its model and others with this release. Much of the company’s blog post about the release of GPT-5.2 focuses on how it has improved on GPT-5.1, which it pushed out after GPT-5 proved to be a major disappointment. According to Axios, OpenAI CEO for applications, Fidji Simo, denied on a press call that GPT-5.2 was in any way a response to the release of Gemini 3.
OpenAI also claims that GPT-5.2 makes advancements in safety, including in how it responds when users show signs of mental distress, and now produces “fewer undesirable responses” in sensitive situations. The company was recently sued for the wrongful death of an 83-year-old woman and her son, who killed her and himself after conversations he had with ChatGPT. It is one of several wrongful death lawsuits the company has been hit with, which have revealed troubling conversations between users and ChatGPT.
