OpenAI’s “code red” response to Google’s Gemini 3 Pro is here. On the same day the company announced a Sora licensing deal with Disney, it unveiled GPT-5.2. OpenAI touts the new model as the best yet for real-world business use. “It is better at creating spreadsheets, creating presentations, writing code, perceiving images, understanding long contexts, using tools, and managing complex, multi-step projects,” OpenAI said.
In a series of 10 benchmarks highlighted by OpenAI, GPT-5.2 Thinking, the most advanced version of the model, outperformed its GPT-5.1 counterpart, sometimes by a wide margin. For example, in AIME 2025, a test that involves 30 difficult math problems, the model achieved a perfect score of 100%, beating GPT-5.1’s already advanced score of perfect 94. He also achieved this feat without relying on tools like web search. Meanwhile, in ARC-AGI-1, a benchmark that tests an AI system’s ability to reason abstractly like a human would, the new system beat GPT-5.1’s score by more than 10 percentage points.
OpenAI claims that GPT-5.2 Thinking is more effective at answering questions in a factual manner, with the company finding that it produces errors 30% less frequently. “For professionals, this means fewer errors when using the model for research, writing, analysis and decision support purposes, making the model more reliable for everyday knowledge work,” the company said.
The new model should also be better in conversation. About the version of the system most users are likely to encounter, OpenAI says: “GPT-5.2 Instant is a fast, powerful tool for everyday work and learning, with marked improvements in information-seeking questions, how-tos and walkthroughs, technical writing, and translation, building on the warmer conversational tone introduced in GPT-5.1 Instant.” »
While it’s probably a stretch to suggest that this is a breakthrough release for OpenAI, it’s fair to say that the company is banking heavily on GPT 5.2. Its big 2025 release, GPT-5, failed to meet expectations. Users complained about a system that generated surprisingly stupid responses and had an annoying personality. The disappointment with GPT-5 was so great that people started demanding that OpenAI bring back GPT-4o.
Then came Gemini 3 Pro, which rose to the top of LMArena, a website where humans evaluate the results of AI systems to vote for the best. Following Google’s announcement, Sam Altman reportedly called for a “code red” effort to improve ChatGPT. Before today, the company’s previous model, GPT-5.1, was ranked sixth on LMArena, with Anthropic’s systems and Elon Musk’s xAI occupying the spots between OpenAI and Google.
For a company that recently signed infrastructure contracts worth over $1.4 trillion in an effort to outpace the competition, this was not a good position for OpenAI. In his memo to staff, Altman said GPT-5.2 would be the equal of Gemini 3 Pro. As the new system rolls out, we’ll see if this is true and what it could mean for the company if it can’t at least match Google’s best.
OpenAI offers three different versions of GPT-5.2: Instant, Thinking and Pro. All three models will first be available to users of the company’s paid plans. Notably, the company plans to keep GPT-5.1 around, at least for a little while. Paid users can continue using the old model for the next three months by selecting it from the old models section.
