OpenAI’s Deep Research: The Most Accurate AI Agent in 2025?

The rivalry for the best and most accurate AI model has never been more fierce. Although DeepSeek’s R1 provided serious competition to multibillion-dollar AI projects like OpenAI’s ChatGPT, the tech giant proves it doesn’t ‘burn’ through those billions for nothing.

On February 2, 2025, OpenAI introduced the latest step forward: deep research, a new AI agent capable of conducting multi-step research on the Internet for complex tasks.

According to OpenAI, deep research can achieve in minutes what would take a human several hours.

Here’s everything we know about potentially the most accurate AI tool so far.

Key Takeaways

OpenAI’s deep research is a new ChatGPT agent for multi-step internet research on complex tasks.
Powered by the o3 model, it uses reasoning and intelligently browses and synthesizes information from text, images, and PDFs.
It takes 5-30 minutes to generate detailed, research-grade reports.
Early tests show record-breaking accuracy, but it can still produce occasional errors.
Deep research is currently available for ChatGPT Pro users ($200/month) with 100 monthly queries.

Table of Contents Table of Contents

Key Takeaways
What Is Deep Research?
What Makes OpenAI's Deep Research Special
Who Needs Deep Research? Early Users Are Impressed
Deep Research Limitations: High Compute Cost & Hallucinations
Is Deep Research Publicly Available?
The Bottom Line: What's Next?
FAQs
References

Table of Contents

Key Takeaways
What Is Deep Research?
What Makes OpenAI's Deep Research Special

Show Full Guide

Who Needs Deep Research? Early Users Are Impressed
Deep Research Limitations: High Compute Cost & Hallucinations
Is Deep Research Publicly Available?
The Bottom Line: What's Next?
FAQs
References

What Is Deep Research?

Deep research is a new agentic feature in ChatGPT, which performs multi-step research on the internet for complex tasks.

Powered by the latest OpenAI o3 optimized for web browsing, it uses reasoning to intelligently and extensively browse text, images, and PDFs across the internet to give you a comprehensive report.

What Makes OpenAI’s Deep Research Special

Deep research’s underlying model o3, optimized for data analysis and web browsing, enables it to use reasoning to search, analyze, interpret, and synthesize massive amounts of information.

The research agent might not give you an instant response. Deep research can take 5 to 30 minutes to complete its analysis, using this time to dive deep into all the information available on the topic.

Thomas Randall, research lead at Info-Tech Research Group, told Techopedia:

“The increased deliberation and time spent fact-checking its output is to be commended. The o3 model family may have some delay while it processes information, but the reliability of the output is that much more improved.”

According to OpenAI, deep research marks a significant step toward a broader goal of developing AGI.

New High at Humanity’s Last Exam

Deep research showed a groundbreaking score of 26.6% – setting a new record for accuracy – at Humanity’s Last Exam.

This test consisted of more than 3,000 multiple-choice and short-answer questions covering more than 100 subjects, from linguistics to rocket science.

OpenAI states, “the model powering deep research showcased a human-like approach by effectively seeking out specialized information when necessary.”

The tool achieved twice the score of o3-mini, and can even perform some tasks that would take PhD experts 10+ hours to do, said Jason Wei, AI researcher at OpenAI.

OpenAI's score at Humanity's Last Exam — OpenAI’s score at Humanity’s Last Exam. Source: OpenAI

Future Implications of Deep Research

Commenting on the implications of deep research assistance, Wei said:

“Deep research can be seen as a new interface for the internet, in addition to being an incredible agent…With a deep research model that browses the web so well, we are entering a world where any information that would take human hours to compile can be synthesized by AI for you in a few minutes…It’s basically like a version of the internet personalized for what you want to know.”

According to Wei, AI research will make the traditional internet search obsolete: “This paradigm will be so powerful that in the future, navigating the internet manually via a browser will be ‘old-school,’ like performing arithmetic calculations by hand instead of using a calculator.”

Who Needs Deep Research? Early Users Are Impressed

Deep research is primarily meant for those who perform intensive knowledge work and require reliable and thorough research in various fields, including finance, engineering, policy, and science.

For example, Derya Unutmaz, biomedical scientist, human immunologist, and Professor at The Jackson Laboratory, found OpenAI’s deep research an “absolute game-changer for scientific research.”

I can finally reveal that I’ve had early access to @OpenAI’s Deep Research since Friday & I’ve been using it nonstop! It’s an absolute game-changer for scientific research, publishing, legal documents, medicine, education-from my tests but likely many others. I’m just blown away!

— Derya Unutmaz, MD (@DeryaTR_) February 3, 2025

Mushtaq Bilal, PhD, Co-founder of Research Kick, believes that soon AI tools like OpenAI’s deep research will do the actual research, and experts will supervise AI and revise the output.

OpenAI just introduced Deep Research, a new AI agent that can conduct complex research using online sources.

Very soon, instead of conducting literature reviews, especially systematic and exhaustive reviews, academics will fine-tuning AI agents like Deep Research for these…

— Mushtaq Bilal, PhD (@MushtaqBilalPhD) February 3, 2025

Ethan Mollick, Associate Professor at The Wharton School, compares ChatGPT’s deep research to an opinionated PhD-level researcher who follows lead.

OpenAI’s deep research is very good. Unlike Google’s version, which is a summarizer of many sources, OpenAI is more like engaging an opinionated (often almost PhD-level!) researcher who follows lead.

Look at how it hunts down a concept in the literature (& works around problems) pic.twitter.com/zAQSBrP2i7

— Ethan Mollick (@emollick) February 3, 2025

However, it can also be beneficial for casual users. For example, the general public, including travelers and shoppers, will greatly appreciate quick and accurate summarization of product, restaurant, and hotel reviews.

Deep Research Limitations: High Compute Cost & Hallucinations

OpenAI admits that its latest development in ChatGPT is very compute-intensive. The longer it researches a query, the more inference computation is required.

Highlighting the high cost of OpenAI’s o3 model, powering deep research, Randall told Techopedia:

“This is to the point where OpenAI has claimed that the o3 models can meet the ‘conventional understanding’ of the AGI benchmark. However, the cost of doing so is not currently economical – the high compute setting costs thousands of dollars per task.”

In other concerns, deep research can sometimes still hallucinate facts or make incorrect inferences, but at a much lower rate than other existing ChatGPT models.

ChatGPT’s new agent can struggle with distilling authoritative information from rumors. There might also be some formatting errors in citations.

Dan Shipper, Co-founder and CEO at Every, who’s been testing OpenAI’s deep research for a few days, said:

“It’s like a bazooka for the curious mind.”

However, he also found some limitations:

It doesn’t always fully cite where a piece of information came from.
There’s no stop button yet, so “if it’s going off the rails, you have to start over.”

“But it’s very clearly a peak into the future of human-AI collaboration for knowledge work,” Shipper concluded.

Is Deep Research Publicly Available?

Deep research is available only for ChatGPT Pro users with up to 100 monthly queries. ‘Plus’ and ‘Team’ users will get access next, followed by ‘Enterprise.’

OpenAI plans to follow its iterative deployment strategy, promising significantly higher rate limits to all paid users when they release a faster, more cost-effective version of deep research powered by a smaller model.

The company plans to release deep research to Plus users in about a month.