Skip to content

OpenAI

Solopreneur Choice

Deep research is an agentic capability in ChatGPT that conducts multi-step research across the web and returns a cited report. It is built for people doing intensive knowledge work in finance, science, policy, and engineering, and for shoppers who need carefully researched purchase recommendations.

SOC 2ISO 27001
370

Overview

Deep research is an agentic capability inside ChatGPT that takes a single prompt and works independently to find, analyze, and synthesize hundreds of online sources into a comprehensive report written at the level of a research analyst. OpenAI built it for people who do intensive knowledge work in finance, science, policy, and engineering, and it can be just as useful for shoppers who want hyper-personalized recommendations on purchases that normally require careful research, such as cars, appliances, and furniture.

What the agent actually does

You give deep research a prompt, and it searches, interprets, and analyzes large amounts of text, images, and PDFs on the internet, pivoting as needed in response to what it encounters. The system is powered by a version of the OpenAI o3 model optimized for web browsing and data analysis, and it was trained on real-world tasks requiring browser and Python tool use through the same reinforcement learning methods behind OpenAI o1. That training taught it to plan and execute a multi-step trajectory, backtracking and reacting to real-time information where necessary. OpenAI positions the ability to synthesize knowledge as a prerequisite for creating new knowledge, and describes deep research as a step toward its broader goal of developing AGI capable of producing novel scientific research.

How a research run works

In ChatGPT, you select deep research in the message composer and enter your query, and you can attach files or spreadsheets to add context. Once it starts running, a sidebar appears with a summary of the steps taken and sources used. A run may take anywhere from 5 to 30 minutes, depending on how deep it needs to go. You can step away or work on other tasks and get a notification when the research is complete. The final output arrives as a report within the chat. At launch, OpenAI said it would add embedded images, data visualizations, and other analytic outputs to these reports in the following weeks.

Where it fits against GPT-4o

OpenAI draws a clear line between deep research and GPT-4o. GPT-4o is ideal for real-time, multimodal conversations, while deep research is aimed at multi-faceted, domain-specific inquiries where depth and detail are critical. The difference, in OpenAI's framing, is between a quick summary and a well-documented, verified answer that can be used as a work product. Every output is fully documented, with clear citations and a summary of its thinking, which makes it easier to reference and verify. The tool is particularly effective at finding niche, non-intuitive information that would otherwise require browsing numerous websites.

Updates since launch

OpenAI has expanded deep research several times. On July 17, 2025, it gained access to a visual browser as part of ChatGPT agent, available by selecting agent mode from the composer dropdown, while the original deep research option remained in the tools menu. On February 10, 2026, OpenAI added the ability to connect deep research to any MCP or app and to restrict web searches to trusted sites for authenticated, industry-standard sources. That update also brought real-time progress tracking, the ability to interrupt a run and refine it with follow-up prompts or new sources, and a revised visual experience for starting, tracking, and reviewing research end to end.

Access and limits

Deep research rolled out in stages. It launched for Pro users, with Plus and Team users next and Enterprise after that. On February 25, 2025, all Plus users gained access, and on February 5, 2025, Pro users in the United Kingdom, Switzerland, and the European Economic Area got it. On April 24, 2025, OpenAI increased usage limits and introduced a lightweight version powered by a version of o4-mini, designed to be more cost-efficient while preserving high quality. Once you reach your limit for the full version, queries automatically switch to the lightweight version.

What to keep in mind

The trade-off is time. A deep research run can take 5 to 30 minutes, so it is not the right tool for a quick lookup. Usage is capped by plan, and the caps have changed more than once, so it is worth checking the current limits for your subscription. Access itself has been staged by plan and region, which means availability depends on where you are and what you pay for. OpenAI has also said that future connections to more specialized data sources, including subscription-based or internal resources, are planned to make output more robust and personalized.

Connecting Deep Research to MCP servers and your apps

Out of the box, deep research reads the public web and any files you upload. Its less obvious capability β€” the one behind searches like "connect deep research to any MCP or app" β€” is that it can also work from your own sources. How you wire that up depends on whether you are working in ChatGPT or building on the API.

In ChatGPT: connected apps. OpenAI's integrations β€” long known as connectors, now simply called apps β€” link services such as Google Drive and SharePoint to your account from Settings β†’ Plugins. When an app supports deep research and is enabled for your account or workspace, it becomes a research source automatically: ask a question, and the resulting report can cite documents from your Drive alongside pages from the open web. Two guardrails are worth knowing. Deep research only ever reads from connected apps; it never uses their write actions, so a research run cannot change or delete your files. And it respects the permissions of the account you connected β€” it cannot surface anything you could not open yourself. Availability varies by plan, region and workspace settings, and not every app supports deep research. For a service with no directory app, ChatGPT on the web can also add a custom MCP server by URL, which deep research can then query on a read-only basis.

Through the API: MCP as a data source. Developers get the same idea in programmable form. Deep research runs through the Responses API, and a request can combine hosted web search with one or more remote MCP servers β€” for instance, an MCP server that exposes your internal document store through simple search and fetch tools. The model plans its sub-questions, queries those private sources alongside the web, and returns a structured, citation-rich report, with every intermediate step exposed for inspection. Because runs can take minutes, they can execute in background mode while your system polls for completion or waits on a webhook. OpenAI's developer cookbook publishes a reference MCP server built for exactly this pattern.

In short: in ChatGPT you connect apps; in the API you connect MCP servers. Either way, deep research stops being limited to the public web.

Frequently asked questions

Can you connect deep research to an MCP server or an app? Yes, in two ways. In ChatGPT, connect an app such as Google Drive or SharePoint (Settings β†’ Plugins), or add a custom MCP server by URL on the web, and supported sources are used automatically in research runs. Through the API, you attach remote MCP servers directly to a Responses API request as data sources.

Which apps can deep research use as sources? Any connected app that supports deep research and is enabled for your account or workspace β€” document stores such as Google Drive and SharePoint are the common examples, alongside authenticated industry data services. Availability depends on your plan, region, workspace settings and the permissions of the account you connected.

Is there an API for OpenAI deep research? Yes. Deep research is available programmatically through the Responses API: you supply a research prompt plus at least one data source β€” hosted web search, file search over your vector stores, or a remote MCP server β€” with code interpreter as an optional extra. Long-running requests can be executed in background mode.

Can deep research change or delete files in my connected apps? No. Deep research uses only the read actions of connected apps and never their write actions. It can search and cite your documents, but a research run cannot modify them.

Core features

  • Runs multi-step web research from a single prompt and returns a comprehensive report with citations
  • Searches, interprets, and analyzes text, images, and PDFs, pivoting as new information appears
  • Shows a sidebar with the steps taken and sources used while the research runs
  • Accepts attached files and spreadsheets to add context to a query
  • Sends a notification when the report is ready, so you can work on other tasks in the meantime
  • Connects to any MCP or app and can restrict web searches to trusted sites (February 10, 2026 update)
  • Supports real-time progress tracking and interruption for follow-up prompts or new sources (February 10, 2026 update)
  • Includes a visual browser through ChatGPT agent mode (July 17, 2025 update)

Use cases

  1. 1Competitive analysis, such as comparing streaming platforms
  2. 2Personalized purchase research on cars, appliances, and furniture
  3. 3Finding niche, non-intuitive information that would require browsing many websites
  4. 4Domain-specific inquiries in finance, science, policy, and engineering that need documented, verifiable answers
  5. 5Offloading time-intensive web research that would otherwise take hours

Strengths and things to check

Strengths

  • Produces fully documented output with clear citations and a summary of its thinking
  • Handles multi-faceted, domain-specific questions that demand depth and detail
  • Works independently, so you can step away and get notified when the report is ready
  • Effective at surfacing niche information that would take many sites to find manually
  • Newer updates add MCP and app connections, trusted-site restrictions, and real-time progress tracking

Things to check

  • Runs can take anywhere from 5 to 30 minutes, so it is not suited to quick lookups
  • Usage is capped by plan, and queries automatically switch to a lightweight version once the full-version limit is reached
  • Embedded images, data visualizations, and other analytic outputs were described as coming in the next few weeks at launch
  • Access has rolled out in stages by plan and region, so availability depends on your subscription and location

Pricing

Deep research usage is capped by plan. As of the April 24, 2025 update, Plus, Team, Enterprise, and Edu users get 25 queries per month, Pro users get 250, and Free users get 5. At launch, Pro users had up to 100 queries per month. Once the full-version limit is reached, queries automatically switch to a lightweight version powered by a version of o4-mini. ChatGPT plan pricing listed on the official pricing page includes Free at $0 per month, Go at $8 per month, Plus at $20 per month, and Pro from $100 per month.

Basic information

Website
openai.com
Added
Oct 10, 2026
Platforms
Web
Chinese & English
Not marked
API
Marked as supported
Self-hosting
Not marked
Best for
Knowledge workers in finance, science, policy, and engineering, along with discerning shoppers who need thorough, cited research on complex purchases.
Information updated
Nov 10, 2026
Website status
Checking
Corrections Β· Privacy Β· Data requests?
Contact us

Features, pricing and availability may change. Always verify details on the official website. Maintained by steven Β· Last updated Nov 10, 2026 β†’

More on this tool

Related tools

All categories