Skip to content
Tools / Deep Research
Deep Research icon

Deep Research

AI research reports with citations

Paid skills start at $0.005. The final price is shown before running. Failed paid calls do not charge.

What is verified

Catalog facts and aggregate ToolRouter calls. Success uses all recorded calls, including caller errors in the total, and is not a controlled benchmark.

1Maintained skills
1Integrated providers
$0.005Paid calls from
7Recorded calls
85.7%Successful calls
2026-05-26Usage updated
2026-04-17Tool updated

Deep Research turns any question into a cited research report by searching and synthesizing multiple web sources in parallel. Cross-references findings, assigns confidence scores, and returns structured intelligence with field-level citations.

Standard web search returns links — you still have to read and synthesize everything yourself. Deep Research does the reading for you. It explores multiple sources, identifies where they agree and disagree, and produces a structured report you can act on. Multiple depth tiers let you trade speed against thoroughness: a quick scan takes under a minute, a full analyst-grade deep dive runs for up to 25 minutes and covers far more ground. Checked 2026-08-26 against this manifest.

What you can do

  • research — run a deep research job on any topic; choose a processor tier from lite (fast summary) to ultra (analyst-grade deep dive with full cross-referencing)

Who it's for

Analysts, researchers, journalists, consultants, and anyone who needs a thorough answer backed by sources rather than a quick summary. Also useful for due diligence, market research, competitive intelligence, and fact-checking.

How to use it

  1. Use research with a specific, detailed question — more detail gives better results
  2. Choose a processor: pro is the default and handles most research tasks well; ultra for the deepest coverage
  3. Add -fast to any tier name (e.g. pro-fast) for faster results at the cost of some depth
  4. Set output_mode to "text" for a readable markdown report, or "auto" for structured JSON with citations
  5. The tool runs asynchronously — retain the returned job ID and poll it through the supported client flow until complete

Getting started

Try research on the global AI agent platform market with the pro processor, then open primary citations. Representative manifest-schema shape: run_id, content, citation_count [redacted]. Source coverage is not guaranteed; timing varies.

Permissions and setup

  • Parallel AI API Key (secret): Optional: use your own Parallel AI key instead of the platform default Official setup
Deep ResearchPricing: paid

Run deep research on any topic. Automatically explores multiple web sources, cross-references findings, and returns structured intelligence with citations. Choose a processor tier to control depth vs speed.

Returns: Structured research findings with field-level citations, confidence scores, and source excerpts. Content is either structured JSON (auto mode) or a markdown report (text mode).
Loading reviews...

Loading activity...

v0.032026-04-17
  • source_domains and exclude_domains now enforce a 10-domain maximum with a clear error message
v0.022026-03-22
  • Added subtitle, expanded description, and agent instructions
v0.012026-03-20
  • Initial release

Copy these instructions to use Deep Research in Claude, ChatGPT, Copilot, and more.

Related Tools

Open Web Search
Web Search icon
Web SearchWeb, news, images & maps — one toolfrom $0.005
6

Frequently Asked Questions

How deep can the research go?

You can choose from `lite`, `base`, `core`, `pro`, or `ultra`, depending on how much depth you need.

Does it cite its sources?

Yes. The report includes citations and confidence scores so you can verify the findings.

Can I limit the sources it searches?

Yes. Use `source_domains` when you want to constrain the research to specific websites.

Is the result a report or structured data?

Either. Use `output_mode: "auto"` for structured JSON with citations or `output_mode: "text"` for a markdown report.