OpenAI's Deep Research is Out. Your Shortcut to Analyst-Level Reports on Any Topic (Includes Citations Too)

The AI war is getting intense. The generative AI giant OpenAI, on Sunday, fired its latest shot in the race by rolling out a ChatGPT tool, called Deep Research. This launch comes right after DeepSeeK-R1, an artificial intelligence newbie, has gone viral and captured everyone’s attention. 

“This is like a superpower: experts on demand. It can go use the internet, do complex research and reasoning, and give you back a report,” said OpenAI chief, Sam Altman in a post on X about Deep Research.“It is really good, and can do tasks that would take hours/days and cost hundreds of dollars.” 

What is OpenAI’s Deep Research?

OpenAI Deep Research is a new AI tool that is designed to help you perform in-depth, multi-step research quickly using ChatGPT. It does google on your behalf and provides you with a comprehensive report on any complex topic, that too at the level of a research analyst.

Technically speaking, Deep Research is an AI agent. It’s more than just a chatbot. A traditional chatbot can only answer simple questions, write text information, or generate images for you. 

But OpenAI Deep Research goes beyond that. It can interact with different online sources and software systems to complete the tasks for you.

Key Features of OpenAI’s Deep Research

It’s important to note that Deep Research by OpenAI is integrated within the ChatGPT. It gives ChatGPT web research skills and keeps you from searching a million different places for your research.

Here are the key features of OpenAI Deep Research:

• Multi-Step Research: You can input complex queries and Deep Research will gather data from various sources, such as text, images, and PDF. 
• Comprehensive Reporting: Deep Research presents its output in the form of a well-documented and well-cited report that summarizes all the findings. This report also explains the reasoning behind each conclusion drawn.
• File Handling: Want to provide additional context for your queries? Deep Research allows you to attach files and spreadsheets for better understanding and better results.
• Optimized Reasoning Technology: Deep Research utilizes a variant of OpenAI's newest o3 model, which is a reasoning technology optimized specifically for online browsing and data analysis.

What Kind of Tasks Is Deep Research Good For?

Deep Research, as the name implies, is a good tool for tasks that require deep online research and analysis. You can use it for tasks that generally include intense knowledge work such as complex science questions or asking for recommendations related to cars and appliances.

Some of the ideal tasks for Deep Research include:

• Conducting detailed financial analysis
• Collecting research findings across multiple disciplines
• Analyzing policy implications and regulatory changes
• Producing comparative studies and literature reviews
• Finding hard-to-locate data that requires cross-referencing 

The Speed of OpenAI’s Deep Research 

Going deeper always takes time. Let’s say you’re a grad student. You’re working on a robotics presentation. You need to learn about the latest developments in autonomous vehicle sensors, compare different technologies, and see what’s up next. A project like this would take you several hours of research, right? 

But OpenAI Deep Research can finish such tasks in 5 to thirty minutes for you, depending on how complex and extensive research a specific task requires. 

Here’s what OpenAI’s chief product officer, Kevin Weil shared at the event in in Washington:

“It can do complex research tasks that might take a person anywhere from 30 minutes to 30 days.”

That means, if you have to spend half an hour or one month to complete a complex research task, OpenAI Deep Research can do it for you in tens of minutes, as it only takes 5 to 30 minutes to come up with a well-cited response from multiple resources.

An Expert-Level Task Example of OpenAI Deep Research

OpenAI stated that domain experts have rated Deep Research high as it automates multiple hours of manual investigation.

OpenAI offers multiple examples of expert-level tasks that show how valuable Deep Research’s insights could be to the end users, saving them many hours of research work. The founders of Deep Research state that you can use this tool to understand niche-specific problems through journals and scientific studies.

Source: OpenAI

For instance, when ChatGPT Deep Research was prompted with a Chemistry question about “the differences between pure- and mixed-gas sorption for glassy polymers, how the dual-mode sorption model can be used to predict mixed-gas sorption behavior in glassy polymers, and what challenges there are in accurately predicting pure- and mixed-gas sorption using the dual-mode sorption model,” Deep Research completed the task efficiently.

The model went on to analyze different sorption models out there, accessed open-source information, identified key issues, retrieved PDFs, and even refined the information before assembling it all into a coherent response. As reported by OpenAI, this process saved about four hours of work.

 

OpenAI’s Deep Research Sets a New Record for Accuracy

OpenAI’s Deep Research has demonstrated a phenomenal score of 26.6% in a benchmark test known as Humanity’s Last Exam.

With this score, OpenAI Deep Research has set a new record for accuracy. This score is much higher than AI models, including the recent viral DeepSeek R1 which only scored 9.5 on Humanity’s Last Exam.

This benchmark test which Deep Research went through included more than 3,000 questions, both multiple-choice and short-answer questions. These questions covered over 100 subjects, such as linguistics, rocket science, and ecology.

As per OpenAI, the reasoning model powering their Deep Research tool exhibited a human-like ability to efficiently seek specialized information when needed.

OpenAI Deep Research’s Performance at Humanity’s Last Exam. 
Source: OpenAI

Limitations of OpenAI Deep Research

Despite all its strengths, OpenAI admits that there are certain limitations to Deep Research. The internal evaluations state that the tool can still “hallucinate” incorrect facts and may find it difficult to differentiate between authoritative sources and what we call rumors. However, these limitations are at a considerably lower rate when compared with other existing ChatGPT models.

Conclusive Remarks

Deep Research is no doubt a great addition to ChatGPT. It gives ChatGPT the ability to surf the web and synthesize comprehensive, analyst-grade reports for you. We can describe it as an autonomous research assistant that is going to transform the way we search for information online and generate insights and reports.