OpenAI’s “deep Research” Can Actually Produce Professional Reports With Citations
Not to be outdone by Deepseek , OpenAI is launching a new Deep Research feature in ChatGPT . This is OpenAI’s newest agent-based AI feature (after Operator ) and builds on the recent trend of making AI more autonomous. According to OpenAI, Deep Research is capable of producing detailed reports at the level of a research analyst. In layman’s terms, it scans and interprets the Internet for you.
Deep Research uses OpenAI’s future o3 reasoning model to perform complex tasks using its time. This feature is currently available to ChatGPT Pro customers (an expensive subscription that costs $200 per month), but will soon be available to ChatGPT Plus and Enterprise users.
How the OpenAI deep learning AI agent works
OpenAI’s Deep Research tool is designed to work independently of you. You give him a detailed hint, after which he asks clarifying questions. It will then go and do its thing in the background. According to OpenAI, a Deep Research period can last anywhere from 5 to 30 minutes, but the company claims it can complete several hours of human-level work in just a dozen or so minutes.
While it’s running, there’s a panel on the right side of the page that shows everything it’s doing in real time. Think of it as the bot’s quotes, but it also explains his “thought process.” It can connect to the Internet, search the Internet, read web pages, and analyze or synthesize vast amounts of information in the form of text, images, and PDFs. All of this requires a bit more computing power, so OpenAI limits Pro users to just 100 requests per month. A smaller and more efficient model will also be introduced in the coming months.
The Deep Research feature is specifically designed for professionals in the fields of science, finance, technology and politics. But OpenAI says it can be equally useful for consumers. OpenAI gave an example of how Deep Research can help drive hyper-personalized research to inform critical purchasing decisions. For example, help in choosing between a car, furniture, household appliances or electronics. Because the tool can synthesize information from thousands of articles and reviews, it can supposedly create a report tailored to your needs.
According to OpenAI, “Deep mining was rated by domain experts as automating hours of complex manual research.”
OpenAI offers many examples where Deep Research insights can be valuable to users, saving hours of research time. The company says it can be used to understand extremely niche and specific problems through scientific studies and journals.
For example, the ChatGPT chemistry assignment asks to “discuss the differences between pure and mixed gas sorption for glassy polymers, how a dual-mode sorption model can be used to predict mixed gas sorption behavior in glassy polymers”, then the model goes on to understand sorption models, access information from open sources, clarifies key issues, opens PDFs, and even refines the model before putting all the content together. According to OpenAI, this task saved 4 hours of time.
OpenAI’s post also highlights similar use cases for Deep Research in healthcare and linguistics, saving five and two hours respectively.
Deep Research also supposedly performed well on Humanity’s Final Exam, an artificial intelligence test that tests expert-level knowledge in more than 100 fields. Deep Research achieved an accuracy of 26.6%, which is the highest for text. In comparison, DeepSeek-R-1 scored 9.4% and GPT-4o scored just 3.3%.
Although Deep Research is based on a reasoning model rather than an LLM, it still uses a language model to operate on input data and generate output text. OpenAI warns that the Deep Research model can still hallucinate and make things up, so it is still better to follow the research results rather than blindly trust them.