Claude Research (2026): Which Plans Get It, How to Turn It On, and Why It Eats Your Limits

The short answer

As of October 20, 2026, Claude Research is on Pro ($20 a month, $17 billed annually), Max (from $100), Team and Enterprise. It is not on Free. It needs web search turned on, and it always runs on Opus 5.5, the model Anthropic bills as its daily driver, no matter which model you picked. There is no research counter. Every run comes out of the same usage pool as your chats and Claude Code, and Anthropic says plainly that research burns that pool faster. That last part is the whole story.

Below: how to turn it on, what it can read, the mobile catch nobody mentions, what it costs you in limits, how it compares to ChatGPT deep research, and the jobs I'd actually spend it on.

What Claude Research actually does

A normal Claude answer with web search runs a search or two and replies. Research runs multiple searches that build on each other, decides what to look into next based on what it finds, explores different angles and works through open questions before it writes. The difference is the loop: it's an agent doing a research pass, not a chatbot with a search box.

It reads two kinds of sources. The public web, and your connected tools: Google Workspace (Gmail, Google Calendar and Google Docs) plus any other connector you've turned on for that chat. One line in Anthropic's help article matters for anyone with sensitive data: during research, Claude uses your connector tools automatically, without asking for approval each time. Turn off the connectors a given question doesn't need before you hit go. (Setting those up for a business is on Claude for Google Workspace.)

Which plans get Research

Free includes web search but not Research, so "is Claude Research free" has a short answer. What Free does get is on is Claude AI free, and the full plan ladder is on Claude AI pricing.

How to turn on Research (and the mobile catch)

1. Turn on web search. Research won't run without it.
2. Click the + button at the bottom left of the chat box and choose Research. A blue indicator shows at the bottom of the chat window.
3. Ask your question. Click the blue indicator again to switch Research off.

If Claude answers without researching, Anthropic's own fix is to say so: "Claude, please use the research tool to..." Blunt, and it works.

Now the catch. If you start Research in the Claude mobile app, it searches the web only and skips your connectors, with no error or other sign that it did. So the "what did the client say about pricing in last month's emails" question you fired off from your phone came back from the public web and nowhere else. Start anything that needs Gmail or Docs on the web or in Claude Desktop. And Desktop has its own limit: Research can't use tools from local MCP servers you've configured there, only connectors.

What the price doesn't tell you: Research has no counter, and it always runs Opus

ChatGPT gives you a number of deep research tasks. Claude doesn't. Anthropic's pricing page says every plan has limits on a rolling five-hour window, paid plans add weekly limits on top, and "there's no fixed message count." Research is subject to those same limits, but Anthropic's help center adds that research sessions can use them up faster because Claude retrieves multiple sources and writes longer answers. The pricing page also reserves the right to limit "model and feature usage" separately.

Then the detail almost nobody has noticed: Research runs on Claude Opus 5.5 (falling back to Opus 4.8) regardless of the model you selected. You might keep your chats on Sonnet to stretch your allowance. Research ignores that. On Anthropic's API price list, Opus 5.5 is $4 per million input tokens and $20 per million output, against $2 and $10 for Sonnet 5.5, so per token it's twice Sonnet's rate. My arithmetic, not Anthropic's. On subscription plans that shows up as a faster-draining bar, not a bill.

On Enterprise it is a bill, because Enterprise is $20 a seat plus usage at API rates. A labeled illustration, since Anthropic publishes no tokens-per-run figure: a research pass that read 1 million tokens of sources and wrote 100,000 would cost $4 + $2 = $6 at Opus 5.5 rates, against $3 if the same work ran on Sonnet 5.5. Your real runs will be bigger or smaller; the ratio is the part that holds. And if you're on Team or Enterprise and Research suddenly stops working, check whether an admin turned off both Opus 5.5 and Opus 4.8 for your role, because Research won't run without one of them.

What this means in practice:

• Don't use Research for lookups. A plain chat with web search answers "what's Acme's current pricing" for a fraction of the allowance.
• Watch Settings > Usage after your first two or three research runs. That tells you what one costs on your plan better than any blog post can, including this one.
• If you hit the wall daily, paid plans can turn on usage credits at standard API rates, or move up a tier. Whether that upgrade pays is the math on is Claude Max worth it. How the five-hour and weekly meters work is on Claude usage limits.

Claude Research vs ChatGPT deep research

Sources: Anthropic's help article and pricing page, and OpenAI's deep research help article (updated the day before), all read October 20, 2026. Neither vendor says which one is more accurate, and I'm not going to pretend a table settles that. If your research lives in Gmail and Google Docs, Claude is the easier fit. If you need a formatted report you can send a client, ChatGPT's export wins. The full ChatGPT side is on ChatGPT deep research for business, and if you're weighing a search-first tool instead, Claude vs Perplexity covers it.

Five jobs worth a research run

Research earns its share of the allowance when the question needs several sources stitched together and you'd otherwise spend an hour in tabs. The ones I'd spend it on:

• Prospect prep before a sales call: the company's recent news, what they've said publicly about the problem you solve, and (on web or Desktop, with Gmail connected) every thread you've already had with them.
• Vendor shortlists: three to five tools for one job, with each vendor's current pricing page as the source. Then check those pages yourself before you buy anything.
• Competitor sweeps: who's offering what you offer in your market and at what published price.
• Meeting prep from your own data: pull a client's history from Gmail, Calendar and Docs into one brief. This is the job ChatGPT can't do as easily from your Google account.
• Regulation or policy questions: what changed and when, with links to the primary source, which you then read.

One rule for all of them: click three of the sources before you act on the answer. A research agent that runs many searches can also stack many small mistakes, and a confident summary of a stale page is still a stale page.

Is Claude Research worth it?

If you're already paying for Pro, it's included, so the only cost is allowance. Use it for the handful of multi-source questions a week that would otherwise eat an afternoon, and keep plain chat for everything else. If a team is going to lean on it, the Team plan's Premium seats (5x a Standard seat) exist for exactly this kind of heavy use; seat math is on the Claude Team plan. The model underneath is covered on Claude Opus 5.5.

The research run is the easy part. What changes a business is what happens after the answer: who reads it, what gets decided, what runs every week without you. If you want the do-it-yourself path for setting Claude up across a small team, that's The Shop ($397, Lesson 1 free). If you'd rather describe your week and get told what to automate first, book a strategy call.

Sources

Read on October 20, 2026: Claude Help Center, "Use research on Claude" (support.claude.com/en/articles/11088861: what Research does, paid plans only, web search requirement, how to turn it on, connectors and Google Workspace, automatic connector tool use, the mobile web-only behavior, no local MCP servers, the steering prompt, same limits but faster use, Opus 5.5 with Opus 4.8 fallback regardless of selected model, the Team and Enterprise admin rule). Claude pricing page (claude.com/pricing: plan prices, Research by plan, usage multiples, the five-hour and weekly limit FAQ, "no fixed message count", feature-usage limits, usage credits, Enterprise seat and usage pricing, Opus 5.5 and Sonnet 5.5 API rates). OpenAI Help Center, "Deep research in ChatGPT" (help.openai.com/en/articles/10500283, via a text reader: usage counter, 30-day allowance reset, research plan, site restriction, connected apps, downloads). Prices in USD. The Opus-to-Sonnet ratio and the token illustration are my arithmetic from the published API rates.