Deep research tools are AI assistant modes that plan a multi-step investigation, run many web searches, read the results and return a long cited report, but each provider names, prices and limits its version differently. This page compares seven of them on identical fields: launch date, what the provider says the tool does, which sources it can read, what it produces, how long a run takes and which plans include it. It covers Claude Research (Anthropic), Gemini Deep Research (Google), Researcher (Microsoft), Le Chat Research mode (Mistral AI), ChatGPT deep research (OpenAI), Perplexity Research (Perplexity) and Grok DeepSearch (xAI). It is written for analysts, researchers and IT teams deciding which tool to use or allow. Tools are listed alphabetically by provider name; the order is never a ranking, and the page draws no verdicts.
Last checked: 2026-11-15.
This is an independent reference page from NHI Mgmt Group based on public provider documentation; it is not sponsored or endorsed by any provider. Deep research features and limits change often, so confirm against each provider's current documentation.
Key takeaways
- The category started in early 2025. Provider pages date ChatGPT deep research to 2 February 2025, Perplexity Deep Research to 14 February 2025, Grok DeepSearch to 19 February 2025 and Le Chat's Research mode to 17 July 2025.
- The workflow is broadly shared. Provider descriptions follow the same pattern: plan the research, search and read iteratively, then write a report with cited sources.
- Sources go beyond the web. Claude Research can use Google Workspace and enabled connectors, Gemini Deep Research can read Gmail, Drive and Chat when the user opts in, Microsoft's Researcher uses work files, emails, meetings and chats, and ChatGPT deep research can connect to apps and MCP servers.
- Plan access differs. Google lists Deep Research on its free tier, Perplexity's free plan includes 1 Research a month, OpenAI marks free deep research as limited, and Anthropic lists Research on paid plans only.
- Microsoft changed its consumer offer. Microsoft says Deep Research is being retired in the consumer Copilot app from 18 August 2026, with Researcher as the reporting experience it points to.
What is a deep research tool?
A deep research tool is an agent mode inside an AI assistant. Instead of answering from one search, it breaks a question into a research plan, runs a series of searches that build on each other, reads the pages it finds, changes direction as it learns, and writes a structured report with citations. Runs take minutes rather than seconds. OpenAI describes deep research as an agent that runs multi-step web research and synthesises many online sources into a report; Anthropic describes Research as running multiple searches that build on each other and deciding what to explore next; Google describes a planning, searching, reasoning and reporting cycle.
Deep research differs from two neighbouring categories. Ordinary web search in a chatbot returns a short answer from a few results. Browser and computer-use agents operate web pages on the user's behalf (clicking, filling forms). OpenAI links the two: its July 2025 update gave deep research a visual browser through ChatGPT agent mode. See Computer-Use and Browser Agents and Personal AI Agents.
Deep research tools at a glance
As of 2026-11-15. Facts are from each provider's own pages listed in Standards and references; "Not stated" means the pages checked did not give the fact.
| Provider | Tool name | First announced | Where it runs | Stated run time | Status |
|---|---|---|---|---|---|
| Anthropic | Research (in Claude) | Not stated by Anthropic on the pages checked (2026-11-15) | Claude on the web, Claude Desktop, Claude Mobile | Not stated by Anthropic on the pages checked (2026-11-15) | Available on paid plans |
| Deep Research (in Gemini) | Not stated by Google on the pages checked (2026-11-15) | Desktop and mobile, 150 countries, 45+ languages, Google Workspace | Not stated by Google on the pages checked (2026-11-15) | Available, including the free tier | |
| Microsoft | Researcher (agent in Microsoft Copilot) | Not stated by Microsoft on the pages checked (2026-11-15) | Microsoft Copilot app, under Agents in Chat | Longer than standard chat by design; no figure stated | Available; consumer Deep Research retired from 18 August 2026 |
| Mistral AI | Research mode (Deep Research, in Le Chat) | 17 July 2025, as a preview | Le Chat | Not stated by Mistral AI on the pages checked (2026-11-15) | Pro plan lists deep research reports |
| OpenAI | Deep research (in ChatGPT) | 2 February 2025 | ChatGPT | 5 to 30 minutes | Available on all plans, limited on Free |
| Perplexity | Research (launched as Deep Research) | 14 February 2025 | Web at launch, with iOS, Android and Mac announced to follow | 2 to 4 minutes; most tasks under 3 minutes at launch | Available, including the free plan |
| xAI | DeepSearch (in Grok) | 19 February 2025 | Grok; X Premium+ at launch | Not stated by xAI on the pages checked (2026-11-15) | Not stated by xAI on the current pages checked (2026-11-15) |
How each tool describes its method
Summaries of each provider's own description. Wording is paraphrased with attribution unless shown in quotation marks.
- Anthropic, Research: runs multiple searches that build on each other, chooses what to explore next based on what it finds and works through open questions systematically. It requires web search to be on and uses connector tools automatically, without asking for approval each time.
- Google, Deep Research: turns the prompt into a multi-point research plan the user can review and refine, browses up to hundreds of websites, shows its thinking as it iterates, and produces a multi-page report.
- Microsoft, Researcher: handles complex, multi-step research tasks and delivers source-cited reports. It may ask clarifying questions first and deliberately spends more time retrieving and analysing than standard chat.
- Mistral AI, Research mode: described at launch as a coordinated research assistant that can plan, clarify the user's needs, search and synthesise.
- OpenAI, deep research: finds, analyses and synthesises many online sources, adjusts its search as it learns, can browse uploaded files, and can plot and refine graphs with Python. Since February 2026 users can track progress and interrupt a run mid-task with follow-up prompts.
- Perplexity, Research: runs many searches, reads a large number of sources, refines its plan as it learns and synthesises findings into a full report.
- xAI, DeepSearch: described at launch as xAI's first agent, built to synthesise key information, weigh conflicting facts and opinions, and end with a concise report.
Sources each tool can read
Source reach matters for both quality and data exposure. As of 2026-11-15.
| Provider | Public web | User files | Email, documents and workspace data | Connectors, apps or MCP | Restrict to chosen sites |
|---|---|---|---|---|---|
| Anthropic | Yes | Not stated by Anthropic on the page checked | Google Workspace: Gmail, Google Calendar and Google Docs | Any connectors turned on for the chat; not local MCP servers configured in Claude Desktop; mobile runs search the web only | Not stated by Anthropic on the page checked |
| Yes, up to hundreds of websites | Yes, uploaded files | Gmail, Drive and Chat, if the user chooses | Not stated by Google on the page checked | Not stated by Google on the page checked | |
| Microsoft | Yes | Work files the user has access to | Emails, meetings and chats the user has access to, respecting existing permissions and policies | Not stated by Microsoft on the page checked | Not stated by Microsoft on the page checked |
| Mistral AI | Yes (search) | Not stated by Mistral AI on the page checked | Not stated by Mistral AI on the page checked | Not stated by Mistral AI on the page checked | Not stated by Mistral AI on the page checked |
| OpenAI | Yes | Yes, uploaded files | Through connected apps | Any MCP server or app, since 10 February 2026 | Yes, trusted sites, since 10 February 2026 |
| Perplexity | Yes | Not stated by Perplexity on the page checked | Not stated by Perplexity on the page checked | Not stated by Perplexity on the page checked | Not stated by Perplexity on the page checked |
| xAI | Yes (described across the corpus of human knowledge) | Not stated by xAI on the page checked | Not stated by xAI on the page checked | Enterprise partners via the API, at launch | Not stated by xAI on the page checked |
What each tool produces
- Anthropic: a research response in the chat; the help article checked does not describe output formats or citations.
- Google: multi-page reports, which can also be generated as an Audio Overview, and interactive content, quizzes and other formats in Canvas.
- Microsoft: reports with visuals, charts and graphs, organised sections and cited sources. Former consumer Deep Research reports can be opened in Microsoft Word and saved.
- Mistral AI: deep research reports, per the Pro plan description.
- OpenAI: a written report in which each claim is cited, with the ability to cite specific passages, a summary of the reasoning, and embedded generated graphs and website images.
- Perplexity: a full report that can be exported to PDF or a document, or converted into a shareable Perplexity Page.
- xAI: a concise, comprehensive report, per the launch description.
Which plans include deep research?
As of 2026-11-15, from each provider's plan or help pages. Plan prices and the full plan comparison are on AI Chatbot Subscription Plans Compared; free tiers are covered on AI Chatbot Free Plans Compared.
| Provider | Free plan | Paid individual plans | Business and enterprise plans | Published quotas |
|---|---|---|---|---|
| Anthropic | Not included | Pro, Max and Ultra | Team and Enterprise | Not stated by Anthropic on the pages checked (2026-11-15) |
| Included | Google AI Plus, Pro and Ultra, with higher general usage limits | Available to Google Workspace users | Not stated by Google on the pages checked (2026-11-15) | |
| Microsoft | Consumer Deep Research retired from 18 August 2026 | Researcher named for Microsoft 365 Premium subscribers | Researcher in Microsoft Copilot at work | Not stated by Microsoft on the pages checked (2026-11-15) |
| Mistral AI | Not stated for Free | Pro lists deep research reports | Not stated by Mistral AI on the pages checked (2026-11-15) | Not stated by Mistral AI on the pages checked (2026-11-15) |
| OpenAI | Limited | Go limited, Plus expanded, Pro maximum | Business and Enterprise limits not stated on the pricing page | Not stated on the current pricing page (see history below) |
| Perplexity | 1 Research a month | Pro and Education Pro: monthly limits for average use; Max: monthly limits for advanced use | Not stated by Perplexity on the pages checked (2026-11-15) | As listed |
| xAI | Not stated by xAI on the current pages checked | X Premium+ at launch; SuperGrok and SuperGrok Heavy named on current pages without feature detail | API access for enterprise partners at launch | Not stated by xAI on the pages checked (2026-11-15) |
OpenAI quota history: the deep research announcement page records, in its 24 April 2025 update, 5 queries a month for Free, 25 for Plus, Team, Enterprise and Edu, and 250 for Pro, with runs switching to a lightweight version powered by a version of o4-mini once the full allowance is used. The current pricing page uses the words limited, expanded and maximum instead of figures, so treat the 2025 numbers as history.
Models behind each tool
- Anthropic: not stated on the Research help article checked.
- Google: Google says the feature launched on Gemini 1.5 Pro, was later improved with Gemini 2.0 Flash Thinking (experimental), and that Gemini 3 improves every stage.
- Microsoft: not stated on the Researcher page checked.
- Mistral AI: not stated for Research mode on the launch page; the same announcement names Magistral for Think mode.
- OpenAI: launched on a version of o3 optimised for web browsing and data analysis, with a lightweight o4-mini based version added in April 2025.
- Perplexity: not stated on the launch page checked.
- xAI: DeepSearch was announced alongside Grok 3; the current model behind it is not stated on the pages checked.
For the current models from each provider, see the hubs for Anthropic Models, Google Gemini Models, Microsoft MAI and Phi Models, Mistral AI Models, OpenAI Models and xAI Models.
Data handling and security for deep research
Deep research reads far more content than a chat answer, and several tools reach into mailboxes and drives. That changes the risk picture in three documented ways:
- Automatic tool use: Anthropic states that Research uses connector tools without asking for approval each time. Any tool can read and act on whatever its connection allows, so scope connectors before enabling them.
- Workspace reach: Google's Deep Research can read Gmail, Drive and Chat when chosen, and Microsoft's Researcher reads work files, emails, meetings and chats within the user's existing permissions. Over-shared files become research input. See the Gemini for Google Workspace Security Guide and Microsoft 365 Copilot Security Deep Dive.
- Untrusted web content: a tool that reads hundreds of pages will meet pages written to manipulate AI readers. OpenAI's option to restrict searches to trusted sites addresses this. See LLM Grooming and AI Search Poisoning.
For enterprise controls, see the ChatGPT Enterprise Security Guide, Claude for Enterprise Security Guide, Gemini Enterprise Security Guide and Perplexity Enterprise Security.
How to choose a deep research tool: factual criteria
- Sources needed: if the research must include internal email or documents, check which tools state workspace or connector access (Anthropic, Google, Microsoft, OpenAI) and whether your organisation has approved that connection.
- Source control: if results must come only from vetted sites, check for a stated site restriction option (OpenAI states one since February 2026).
- Volume: if you need many runs a month, check for published quotas (Perplexity publishes free and enterprise figures) or test the limit on the plan you hold.
- Output format: check for the format you need, such as PDF export (Perplexity), Word (Microsoft), audio summaries (Google) or embedded charts (Microsoft, OpenAI).
- Time per run: stated run times range from 2 to 4 minutes (Perplexity, at launch) to 5 to 30 minutes (OpenAI); plan work around them.
- Verification effort: every tool cites sources, but citations still need to be opened and checked before a report is relied on.
Practitioner checklist
- Open and check a sample of citations in every deep research report before it informs a decision or is published.
- Decide which connectors (mail, drive, calendar, MCP servers) may be used with deep research, and disable the rest at the organisation level.
- Fix over-shared files before enabling workspace-aware research, since the tool inherits the user's access.
- Use trusted-site restrictions where the tool offers them for regulated or high-stakes research.
- Keep deep research for work data on managed business plans, not personal free accounts; see the Blocking Personal ChatGPT, Claude and Gemini Accounts at Work.
- Record retirements and changes, such as Microsoft's consumer Deep Research retirement on 18 August 2026, and update guidance and saved-report procedures.
- Re-check quotas on the live plan pages; OpenAI's published figures from 2025 have been replaced by tier wording.
Standards and references
- Anthropic: Using Research on Claude (accessed 2026-11-15)
- Anthropic: Claude plans and pricing (accessed 2026-11-15)
- Google: Gemini Deep Research overview (accessed 2026-11-15)
- Google: Gemini subscriptions (accessed 2026-11-15)
- Microsoft Learn: Researcher agent in Microsoft Copilot (accessed 2026-11-15)
- Microsoft Support: Deep Research is being retired in the Copilot app (accessed 2026-11-15)
- Mistral AI: Le Chat dives deep (accessed 2026-11-15)
- Mistral AI: Pricing (accessed 2026-11-15)
- OpenAI: Introducing deep research (accessed 2026-11-15)
- OpenAI: ChatGPT pricing (accessed 2026-11-15)
- Perplexity: Introducing Perplexity Deep Research (accessed 2026-11-15)
- Perplexity Help Center: Which Perplexity subscription plan is right for you (accessed 2026-11-15)
- xAI: Grok 3 announcement (accessed 2026-11-15)
- xAI: Grok help and FAQ (accessed 2026-11-15)
Related NHI Mgmt Group resources: AI Chatbot Free Plans Compared, AI Chatbot Subscription Plans Compared, Computer-Use and Browser Agents, ChatGPT vs Claude vs Gemini vs Grok (Apps), Frontier AI Models Compared.