A few days after Google sent an internal red alert for artificial intelligence capabilities, OpenAI responded with the launch of GPT-5.2, a model described as the companys strongest to date.
This is a new generation of AI, optimized for the business environment, focusing on professional work and long-term operational actors.
According to OpenAI, GPT-5.2 is superior in speed, reasoning ability and performance in performing a series of in-depth tasks such as summarizing documents, writing code, troubleshooting, building presentations or organizing complex data.
The company said that ChatGPT Enterprise users currently save 40-60 minutes a day, equivalent to 10 hours of work per week, and this figure is expected to increase further with GPT-5.2.
The model is divided into 3 versions: Instant (quick response), Thinking (deep debate) and Pro (highest quality for technical problems).
GPT-5.2 is already available on paid ChatGPT packages, and also appears on API with the names gpt-5.2, gpt-5.2-chat-latest and gpt-5.2-pro.
In terms of performance, GPT-5.2 achieved impressive results in GDPval, a test that simulates 44 practical professions.
The Thinking version alone exceeded or was equal to a human expert in 70.9% of the tasks, double the GPT-5 achievement.
In SWE-Bench Pro for programming, the model sets a new record for outstanding in troubleshooting, deploying features and handling comprehensive technical tasks.
Another highlight is the huge ability to analyze textbooks. GPT-5.2 tracks hundreds of thousands of keywords and still maintains almost absolute accuracy, even when important information is buried deep in hundreds of pages of documents. The model also significantly reduces hallucinations compared to the previous generation.
The ability to use external tools of GPT-5.2 has also been greatly improved. On Tau2 test, the model achieved 98.7% accuracy in multi-step customer support scenarios such as booking air tickets again, finding luggage, arranging a hotel or handling medical requests, situations where the old model was often interrupted in between.
GPT-5.2 is also stronger in solve problems, scientific reasoning and technical imaging. For postgraduate scientific questions, the model achieved an accuracy of more than 92%, and at the same time set a record in many specialized toanistic formats.
Researchers say the model can also propose evidence in statistical theory for human consideration.
The launch of GPT-5.2 comes amid intense AI competition. After Google launched Gemini 3 and Anthropic launched Claude Opus 4, OpenAI CEO Sam altman asked the entire company to focus on improving model quality, suspending side plans, including advertising.
With GPT-5.2, OpenAI expects to improve enterprise AI productivity standards, provide more reliable models, less errors and serve more complex processes in practice.
