OpenAI Launches GPT-5.2 as It Navigates ‘Code Red’

OpenAI has introduced GPT-5.2, its smartest artificial quality exemplary yet, with show gains crossed writing, coding, and reasoning benchmarks. The motorboat comes conscionable days aft CEO Sam Altman internally declared a “code red,” a company-wide propulsion to amended ChatGPT amid aggravated contention from rivals.

“We announced this codification reddish to truly awesome to the institution that we privation to marshall resources successful 1 peculiar area, and that's a mode to truly specify priorities,” said OpenAI’s CEO of applications, Fidji Simo, successful a briefing with reporters connected Thursday. “We person had an summation successful resources focused connected ChatGPT successful general.”

Simo denied that OpenAI had moved up GPT-5.2’s motorboat successful airy of its codification red, claiming the institution has been moving connected this model’s merchandise for months. However, she said the further resources astir ChatGPT person been “helpful.”

While OpenAI’s models and products were considered champion successful people erstwhile ChatGPT launched successful 2022, that’s nary longer a settled matter. The startup present faces an array of worthy challengers, possibly nary much threatening than Google, whose precocious launched Gemini 3 exemplary was received good by the tech industry. Google’s Gemini app has grown astatine an awesome complaint implicit the past year, present with much than 650 cardinal monthly progressive users, compared to OpenAI’s 800 cardinal play progressive users. That unit has forced OpenAI to rein successful immoderate of its astir ambitious projects, including its enactment connected introducing ads to ChatGPT, and to refocus connected improving its halfway exertion and products.

Much similar the company’s caller exemplary launches, GPT-5.2 is shipping arsenic bid of models: Instant, which responds faster and is amended for information-finding; Thinking, which excels astatine coding, math, and planning; and Pro, the astir almighty tier of OpenAI’s models that delivers higher accuracy connected hard questions.

OpenAI calls GPT-5.2 its champion exemplary yet for mundane nonrecreational use. GPT-5.2 Thinking notched the highest scores to day connected GDPval, an OpenAI benchmark that compares show betwixt AI models and quality professionals crossed 44 real-world occupations. The institution says the exemplary bushed quality professionals successful implicit 70 percent of tasks, and completed them 11x faster.

OpenAI’s post-training pb Max Schwarzer says that GPT-5.2 should besides connection a important betterment astir hallucinations successful ChatGPT. The institution says GPT-5.2 Thinking hallucinated 38 percent little than GPT-5.1 connected benchmarks measuring answers to factual questions.

The institution is bringing GPT-5.2 to some ChatGPT users and developers connected OpenAI’s API product. OpenAI says the caller bid of models “brings wide gains crossed mundane and precocious usage cases.”