Přeskočit na obsah

The best AI deep research tools compared

AI deep research tools read dozens of sources and write a cited report. ChatGPT, Gemini, Claude, Perplexity and nerix compared on price, limits and checking.

Tým nerixuZveřejněno 5 min čtení

V tomto návodu

AI deep research tools are assistants that take a question, plan their own searches, read many pages and come back some minutes later with a long report that cites its sources. The main ones in 2026 are ChatGPT, Google Gemini, Claude and Perplexity, plus workspaces such as nerix that run research as a background task. They differ most in price, how many reports you get a month, and how easy it is to check what they wrote. This guide compares them, and shows how to get a report you can rely on.

Prices checked in November 2026. We could open the official pages for Gemini and Claude. ChatGPT's page opened but showed no prices in the text we could read, and Perplexity's pricing page refused our request, while third party sites disagree about their limits, so for those two we give no numbers: check ChatGPT's and Perplexity's own pages.

The tools side by side

Tool Best for Price Free plan Key limit
ChatGPT Long, structured reports Paid plans, check the live page Limited use Allowances are described in words, not numbers
Gemini Reports inside a Google account AI Pro 21.99 euros a month (France page) Yes, limited Limits depend on compute and reset every five hours
Claude Careful reading and writing Pro $17 a month billed yearly, $20 monthly No Research is not in the free plan
Perplexity Fast answers with visible sources Paid plans, check the live page Yes, limited Monthly limits by plan, check the live page
nerix Research you cap and approve Free credits, then Plus 15 euros Yes, 3,000 credits a month Younger product

ChatGPT

ChatGPT's deep research mode plans a search, reads many sources and returns a long report. OpenAI's pricing page describes the allowance in words, not numbers: limited access on Free and Go, expanded on Plus and the maximum on Pro. Figures that circulate online date from 2025 and may be out of date, so look at the counter inside the app.

Good for: well structured, long reports, and people who already use ChatGPT daily.

Falls short: you cannot easily tell in advance how many reports a plan gives you.

Google Gemini

Gemini's Deep Research is part of the Gemini app. On Google's French pricing page, Gemini is free with a Google account, Google AI Plus is 4.99 euros a month, Google AI Pro is 21.99 euros and Google AI Ultra is 99.99 or 219.99 euros. Google does not give a number of reports. It says limits depend on compute and refresh every five hours until a weekly limit is reached, with Plus at twice the free limits and Pro at four times. Prices differ by country.

Good for: people who live in Google Workspace or want a generous Pro tier.

Falls short: the lack of a stated number of reports makes budgeting hard.

Claude

Claude's Research is included in Pro, Max and Team, but not in the free plan. Anthropic lists Pro at $17 a month billed annually ($200 up front) or $20 billed monthly, and Max at $100 a month. Team standard seats are $20 per seat a month billed annually, or $25 monthly.

Good for: careful reading of long sources and clear writing.

Falls short: nothing for free, and the price is in dollars.

Perplexity

Perplexity built its name on answers with visible sources, and its Deep Research mode extends that to longer reports. Perplexity's help centre describes monthly limits on paid plans and a highest tier called Max. Pro is widely reported at about $20 a month, but we could not confirm that on Perplexity's own pages, and reports on how many deep research queries each plan gets disagree, so read the live plan page before you pay.

Good for: quick, source heavy answers and a fast habit of clicking through to the page.

Falls short: unclear and changing limits.

nerix long tasks

We make nerix, so weigh this section with that in mind. nerix runs research as a long task: you describe the result you want, pick the kind of task (research is one of six, alongside general, document, website, compliance check and case file) and choose Economy, Standard or Deep for how much care it takes. Before it starts you see an estimate of the cost. You set the most credits it may use, such as 500, 2,000, 10,000 or your own number, and that is a hard stop: the task ends there and keeps what it has made so far. Ten thousand credits are one euro.

It runs on nerix's servers, so you can close the tab. When it needs you, a question or an approval, it appears under Waiting for you and in the bell. If you turn on Show me the plan before it starts, it writes the plan and waits for your OK. A task can read only what the person who started it may read: the workspace's files, the web when allowed, and what you attach. Reports link each claim to where it came from.

The free plan includes 3,000 credits a month, which is 30 cents worth, enough to try it. Plus is 15 euros a month with 100,000 credits.

Good for: people who want a cost ceiling, a plan to approve and research that can use their own files.

Falls short: it is a younger product, and long Deep runs use real credits.

How to get a report you can rely on

  • Ask a narrow question. "Compare three payroll tools for a team of ten in the Netherlands" beats "tell me about payroll".
  • Say who it is for and what it must contain. Audience, length and format change the result.
  • Name the sources you trust, and the ones to ignore, such as sponsored lists.
  • Ask for a short plan first. If the plan looks wrong, you save the whole run.
  • Open the links behind the claims that matter. Check the figure on the source page.
  • Check dates. Pricing and rules change, and a report can quote an old page.

How to choose

If you already pay for ChatGPT or Gemini, use the research mode you have and see whether the allowance is enough. If you want careful reading and writing, try Claude Pro. If you want quick answers with visible sources, try Perplexity. If you want a spending limit, a plan to approve and research that can use your own documents, try long tasks in nerix, and use chat with sources for quick questions. Run the same question in two tools and compare the sources they chose.

The short version

Deep research saves hours when the question is broad and the sources are open. The big four differ mostly in price and limits that are hard to read, so check the live page. Ask narrow questions, review the plan, open the links behind the key claims, and set a cost limit when the tool allows one.

Na co se lidé ptají

How long does a deep research report take?
Usually several minutes, and sometimes much longer for a broad question. That is why the good tools run in the background and tell you when they are done, instead of making you wait on the page.
Can I trust a deep research report?
Trust it as a first draft with sources, not as a finished fact. Open the links behind the claims that matter, check dates and numbers, and be careful with topics where the sources are thin or one sided.
Is deep research worth paying for?
If you research for a living, or often need a briefing on a market, a supplier or a regulation, yes, because it saves hours. If you need a quick fact, an ordinary chat with web search is cheaper and faster.
Can deep research read my own documents as well as the web?
Some tools can, some only read the web. If your question depends on your own files, pick a tool that can use them, and check what it is allowed to read.

Sdílet tento návod

V nerixu

Vyzkoušejte to na vlastní práci.

nerix napíše první verzi, odpovídá z vašich dokumentů se stránkou, ze které každá věta pochází, a drží práci týmu na jednom místě. Začněte s bezplatným tarifem.

Čtěte dál