OpenAI has released a new tool called deep research. It is an AI agent designed to handle complex research tasks by browsing the web and compiling detailed reports. The tool is now available to users subscribed to the $200 per month Pro tier.
Deep research does not work like a typical chatbot. Instead of providing quick answers, it takes between five and 30 minutes to generate a report. During this time, the agent reads hundreds of sources, pulling together information from across the internet. It produces a result that OpenAI describes as similar to what a research analyst might create.
The tool is built on a version of OpenAI’s upcoming o3 model. That model has been optimized for web browsing and data analysis. The company claims deep research scores highly on benchmarks designed to test complex problem solving. On the Humanity’s Last Exam, a test meant to challenge AI across a range of subjects, the agent scored 26.6 percent. While that might not sound high, it is well above the 9.3 percent scored by the standard o3 model. OpenAI says this shows the agent is better at synthesizing knowledge from the web.
How the agent works and who can use it
To use deep research, a Pro subscriber opens a chat in ChatGPT and selects the deep research option. Then the user types a question. The user can upload files or spreadsheets to add context. The agent then searches the web, analyzes the information, and produces a report. The user receives a notification when the report is ready.
The tool is useful for people who need thorough research. That might include professionals in finance, science, policy, or engineering. OpenAI suggests it can also help with buying decisions, such as comparing cars or appliances. But the company admits the tool is not perfect. It can make mistakes, sometimes called hallucinations. It may also struggle to distinguish authoritative sources from rumors. Users are encouraged to verify facts using primary sources.
Deep research handles text input and outputs text. It cannot yet analyze images or audio. But OpenAI says it plans to add those features later. The tool can already work with uploaded files and render JavaScript for dynamic web pages.
OpenAI plans to bring deep research to Plus and Team subscribers after the Pro launch. The company says it will expand the tool to mobile and app versions in the future. The tool currently covers public web content. It cannot access paywalled content or other restricted material.
Comparing deep research to other AI tools
Deep research enters a field that already has several similar tools. Google has a feature called Deep Research that uses its Gemini AI to create reports. Perplexity offers its own research agent. The startup Hebbia also provides an AI tool designed for complex financial and legal research. Each of these tools has a similar goal. They all try to automate the process of gathering and analyzing information from the web.
OpenAI’s version distinguishes itself with the depth of its analysis and the time it takes. The company says the agent is trained to be a careful researcher. It learns from real browsing behavior. Users who try it say the results look like a well documented report with citations. The agent also shows its thinking process, letting users see how it arrived at its conclusions.
The launch marks another step in OpenAI’s push to build agents that can act independently. Agents are a major focus for the company. Earlier this year, OpenAI released Operator, a tool that can perform actions in a web browser. The company says deep research is one part of a larger plan to give ChatGPT the ability to handle long running tasks. OpenAI has hinted at more agent products later this year.
For now, deep research gives Pro users a powerful new way to gather information. The tool is available today. OpenAI expects to improve it over time. As with any new AI system, users should check the results carefully before relying on them. If you want to stay updated on AI tools and their impact, visit {$link_text} for more analysis and news.
