OpenAI on Thursday announced its most advanced artificial intelligence model, GPT-5.2, calling it its best product yet for everyday professional use.
OpenAI says the model is better than previous models at creating spreadsheets, building presentations, recognizing images, writing code, and understanding long contexts. It will be available starting Thursday within OpenAI’s ChatGPT chatbot and its application programming interface (API).
This announcement comes a few weeks after OpenAI announced its GPT-5.1 model. with rival humans google announced a new model last month, prompting OpenAI to declare a “Code Red” effort to improve ChatGPT and set aside other projects.
It’s all part of a high-stakes battle between big tech companies to build the most widely used models as consumers and businesses increasingly incorporate AI into their daily lives and workflows. As OpenAI seeks to justify a $500 billion valuation and more than $1.4 trillion in planned spending, it hopes the GPT family of models can define its future.
“The reason we announced this code red is to really signal to the company that we want to dedicate resources in certain areas, and this is a way to really define priorities and define what’s OK to de-prioritize,” he told reporters at a Thursday briefing. “We’ve seen more resources focused on ChatGPT in general, and I think that’s helped us release this model, but that’s not the reason it’s being released this week in particular.”
OpenAI CEO Sam Altman told CNBC on Thursday that Google’s release of the Gemini 3 model did not have as much of an impact on the company’s metrics as initially feared. He said he expects OpenAI to end Code Red by January.
“When a competitive threat arises, you want to focus on it and respond quickly,” Altman said.
According to OpenAI, GPT-5.2 will be available in Instant, Thinking, and Pro versions. OpenAI says Instant is faster at writing and searching for information, Thinking is better at structured tasks like coding and planning, and Pro provides the most accurate answers to difficult questions.
The company says its model tops industry benchmarks such as SWE-Bench Pro, which evaluates agent coding performance, and GPQA Diamond, a graduate-level scientific reasoning benchmark. On GDPval, an assessment released by OpenAI earlier this year, GPT-5.2 outperformed or tied the industry’s top professionals on 70.9% of clearly specified tasks, the company said.
“We’ve been working on this for months,” Simo said. “We are proud to be able to release models quickly, but this particular integration has been in the works for some time.”
Anthropic’s latest model, Opus 4.5, scored higher than GPT-5.2 on SWE-Bench Verified, a test set that evaluates the software coding capabilities of AI systems. OpenAI told reporters that the benchmark is less “stain-resistant, difficult, diverse, and industry-relevant” than SWE-Bench Pro.
OpenAI was founded as a research institute exactly 10 years ago, but after the launch of ChatGPT in 2022, it has become one of the fastest growing commercial entities in the world. More than 800 million people now use the company’s chatbots every week.
WATCH: AI pressure cooker: OpenAI, Nvidia and Google all come into play
