According to the company, GPT-5.2 delivers improved performance in building spreadsheets and presentations, interpreting images, writing software and managing long, complex inputs. The model is being added immediately to ChatGPT and to OpenAI’s API for developers.
The rollout comes only weeks after OpenAI released GPT-5.1 and as the company faces intense competition from other major AI developers, including Anthropic and Google. Their recent model updates pushed OpenAI to refocus internal resources on advancing ChatGPT, a move executives described as a “code red” prioritization effort.
Fidji Simo, OpenAI’s chief executive for applications, said the internal directive was meant to concentrate the company’s attention on a single priority. “It was a way to clarify what should come first and what could wait,” she told reporters, adding that the new model’s launch timing was not dictated by that shift in emphasis.
OpenAI CEO Sam Altman said in a separate interview that Google’s release of its Gemini 3 model ultimately had less of an effect on OpenAI’s user metrics than the company initially feared. He said he expects the firm to conclude its “code red” period by January.
GPT-5.2 will be offered in three variants. Instant is designed for rapid responses and straightforward writing tasks; Thinking is aimed at complex, structured work such as coding or planning; and Pro is intended for scenarios requiring the highest accuracy.
OpenAI said GPT-5.2 leads several key benchmarks, including SWE-Bench Pro, which evaluates autonomous coding systems, and GPQA Diamond, a test of advanced scientific reasoning. In its GDPval evaluation — an internal measure released earlier this year — the company said the model matched or outperformed top industry professionals on more than 70% of well-defined tasks.
“These upgrades have been in development for many months,” Simo said, noting the rapid pace of model releases but emphasizing that the integration behind GPT-5.2 was a long-running effort.
Anthropic’s newly released Opus 4.5 outperforms GPT-5.2 on SWE-Bench Verified, a separate coding benchmark, but OpenAI said that test is narrower and less representative of real-world use than SWE-Bench Pro.
OpenAI, founded ten years ago as a research lab, has grown into one of the world’s most widely used AI developers since ChatGPT’s debut in 2022. The company says more than 800 million people use its chatbot each week.