OpenAI has presented GPT-5.2, its smartest synthetic intelligence type but, with efficiency beneficial properties throughout writing, coding, and reasoning benchmarks. The release comes simply days after CEO Sam Altman internally declared a “code purple,” a company-wide push to support ChatGPT amid intense festival from competitors.
“We introduced this code purple to truly sign to the corporate that we need to marshal sources in a single specific space, and that’s the reason a approach to truly outline priorities,” stated OpenAI’s CEO of packages, Fidji Simo, in a briefing with journalists on Thursday. “We’ve got had an building up in sources interested by ChatGPT on the whole.”
Simo denied that OpenAI had moved up GPT-5.2’s release in gentle of its code purple, claiming the corporate has been running in this type’s unencumber for months. On the other hand, she stated the extra sources round ChatGPT had been “useful.”
Whilst OpenAI’s fashions and merchandise had been thought to be best-in-class when ChatGPT introduced in 2022, that’s not a settled subject. The startup now faces an array of worthy challengers, possibly none extra threatening than Google, whose lately introduced Gemini 3 type used to be won neatly via the tech business. Google’s Gemini app has grown at an excellent charge during the last 12 months, now with greater than 650 million per 30 days energetic customers, in comparison to OpenAI’s 800 million weekly energetic customers. That force has pressured OpenAI to rein in a few of its maximum bold initiatives, together with its paintings on introducing advertisements to ChatGPT, and to refocus on making improvements to its core era and merchandise.
Just like the corporate’s contemporary type launches, GPT-5.2 is delivery as a chain of fashions: Quick, which responds quicker and is best for information-finding; Pondering, which excels at coding, math, and making plans; and Professional, probably the most robust tier of OpenAI’s fashions that delivers upper accuracy on tough questions.
OpenAI calls GPT-5.2 its ultimate type but for on a regular basis skilled use. GPT-5.2 Pondering notched the best possible rankings up to now on GDPval, an OpenAI benchmark that compares efficiency between AI fashions and human pros throughout 44 real-world occupations. The corporate says the type beat human pros in over 70 p.c of duties, and finished them 11 instances quicker.
OpenAI’s post-training lead Max Schwarzer says the brand new unencumber will have to additionally be offering a considerable relief in hallucinations. The corporate says GPT-5.2 Pondering hallucinated 38 p.c lower than GPT-5.1 on benchmarks measuring solutions to factual questions.
The corporate is bringing GPT-5.2 to each ChatGPT customers and builders on OpenAI’s API product. OpenAI says the brand new sequence of fashions “brings transparent beneficial properties throughout on a regular basis and complicated use circumstances.”
