The launch followed a week of Sam Altman’s so-called code red, which was his way of admitting ChatGPT needed a brain tune-up after Google’s Gemini started beating it at expert reasoning, logic puzzles, tricky maths and spotting pictures. The new model is meant to be better at science, coding, and long-context work while fixing some of the howlers that made users wonder whether ChatGPT had dropped out of school.
OpenAI’s applications chief Fidji Simo told hacks and hackettes: “We designed 5.2 to unlock even more economic value for people” and said it was stronger at building spreadsheets, churning out presentations, writing code and stringing together multistep projects. Simo, who once ran Instacart, has been brought in to turn the lab into a profitable global business. However, the jury is still out on whether another productivity upgrade will pull in the enterprise crowd.
Gartner Analyst Arun Chandrasekaran told The Wall Street Journal that OpenAI’s model demonstrated better reasoning and coding while keeping pace with Google and Anthropic in handling mixed inputs, such as audio and video. He added that the bump in knowledge work could help ChatGPT appear more credible as an office sidekick rather than a chatbot that occasionally forgets how numbers work.
Constellation Research’s Ray Wang called GPT-5.2 a decent response to Gemini, although not enough to halt its momentum. He said “what OpenAI did was make it easier to create office productivity tools” while noting that Gemini remained “more integrated.”
The model’s image generation was conspicuously absent from OpenAI’s list of triumphs. Google’s Banana Nano has been giving it a bloody nose since August, and Altman himself flagged pictures as a priority in his code-red memo, though there was no sign of it in this week’s reveal.
Box boss Aaron Levie said GPT-5.2 showed “a huge jump” over the 5.1 model from November, especially for bread-and-butter tasks like spreadsheets and slide decks. OpenAI claimed the model matched or beat industry professionals on 70.9 per cent of knowledge work tasks in its GDPval metric, which covers 44 occupations across manufacturing, professional services, healthcare and finance.
Paying ChatGPT users began seeing access on Thursday. The company stressed the release had been on the roadmap for months and was not a panic move. Simo said, “Code red, just to put things in perspective, that’s not an uncommon thing,” before adding that more resources were now pointed at ChatGPT, although that was not why 5.2 landed this week.
The first GPT-5 launch in August had arrived with a thud as punters gleefully shared screenshots of the bot mangling basic maths and making a dog’s breakfast of North America’s outline. OpenAI now needs 5.2 to prove it can still lead the pack.
Disney lobbed in a $1 billion investment and agreed to license its characters for ChatGPT and Sora, which was a reminder that the entertainment giants still want a piece of whatever comes next, even if the underlying models are fighting for relevance.