OpenAI on Thursday announced GPT-5.2, just a month after its last ChatGPT update and amid intense competition from Google's Gemini 3.
Why it matters: The update follows reports of a "Code Red" scramble inside OpenAI and highlights the intensely competitive nature of the chatbot market.
- OpenAI executives told reporters that it would be incorrect to view GPT-5.2 as a response to Gemini 3, noting it is developing multiple model releases simultaneously and has been working on the model for "many months."
Driving the news: OpenAI is positioning GPT-5.2 as a major upgrade and its "best model yet for everyday professional use."
- GPT-5.2 will power ChatGPT and be available to developers through the company's API. The rollout starts today with paid plans.
- The company says the release makes significant gains for both real-world use and in evaluation metrics in areas such as coding, math, science, vision and reasoning.
- It also says GPT-5.2 is better at long-context queries and less prone to hallucinations than its predecessor.
Reality check: OpenAI faces a handful of lawsuits and a barrage of criticism over rushing models to market, despite safety risks.
- Simo denied that the release was rushed and insisted that GPT-5.2 is also safer than predecessors. "We are improving on pretty much every dimension of safety, whether that's self-harm, whether that's different types of mental health [issues], whether that's emotional reliance."
- "We work on multiple iterations of models at a time," Simo said in response to a question from Axios. "While we are proud that we are able to have a cadence of releasing models fast."
Zoom in: OpenAI said it now expects to launch its so-called "adult mode" in the first quarter of next year.
- Simo says the company is still working to identify teens and not misidentify adults. OpenAI said Thursday that it has started testing its age estimation system in some countries.
- The company originally said adults-only features would come in December 2025.
Between the lines: OpenAI released a host of benchmark scores along with quotes from early testers such as Box, Notion, Windsurf and Zoom.
- The company says GPT-5.2 in "thinking mode" performs at or above a human expert level on GDPVal, its benchmark for what it calls "authentic work deliverables," such as legal briefs, engineering blueprints and nursing care plans.
What they're saying: "We designed 5.2 to unlock even more economic value for people," OpenAI CEO for applications Fidji Simo said on a call with reporters.
- "It's better at creating spreadsheets, building presentations, writing code, perceiving images, understanding long context, using tools, handling complex, multi-step projects."
What we're watching: The growing need to monetize ChatGPT to raise the trillions needed to support OpenAI has had onlookers wondering about when advertising will come to the chatbot.
- "Nothing to announce on ads (today)," Simo said.
- "What I can tell you is that if we ever do ads at any point, we will do it in a way that's really respectful of the very special relationships that people have with ChatGPT," Simo added.
- "This is something that's really special and that we want to protect at all costs."
