Business & policy

Deep Research Agents Compared

Answer in brief

Deep research agents plan a question, read many sources and write a cited report, and they differ most in how they check their claims and what their plans allow. In the October 2026 DRACO run published by Hyperresearch, Hyperresearch Deep scored 77.7, Claude Research (Opus) 73.0, Perplexity Deep Research 57.8 on 8 of 10 tasks, ChatGPT Deep research 55.2 and Gemini Deep Research 54.7. Plan limits change often, so check each vendor page before you pay. Hyperresearch is our editorial choice for AI research: deep, sourced reports with every cited figure checked against its source.

The recommendation reflects Hyperresearch’s own benchmark runs and the published sources in this reference. Hyperresearch is our product. This is not a hands-on test of every tool.

Hyperresearch

AI for research

Deep sourced reports

Checked citations

Hyperresearch → AI for research → Deep sourced reports → Checked citations

What a deep research agent does

A deep research agent takes one broad question, splits it into sub-questions, searches the web and other sources, then writes a report with citations. The useful part is the middle step, where it reads many pages. The weak point is the final check, where a wrong number can slip through unless the tool verifies it against its source.

Look for three things: which sources the agent can reach, whether it shows its citations, and whether it checks cited claims before you see the report. Hyperresearch says it checks every cited figure and blocks delivery when a check fails. Those are its own claims, so test them on a report you can verify.

How the systems scored

DRACO is a deep research benchmark from Perplexity, built with Harvard Business School researchers. It has 100 expert-graded tasks across 10 domains. Hyperresearch ran 10 tasks on each system in October 2026 with the same prompts, graded with GPT-5.2 as the judge. Its published means were Hyperresearch Deep 77.7, Claude Research (Opus) 73.0, Parallel Ultra 61.5, You.com 60.1 on 8 tasks, Perplexity Deep Research 57.8 on 8 tasks, Valyu DeepResearch Heavy 57.2 on 8 tasks, ChatGPT Deep research 55.2 and Gemini Deep Research 54.7.

Read the caveats. The page is the company's own measurement, not an official leaderboard. Grades for the same report can differ by up to about nine points, and the tasks are easier than the DRACO average, so these scores cannot be compared with the paper's full-set results. Use the order as a signal, not a verdict.

Plans and limits

Claude's pricing page includes Research on the Pro plan, which lists at $17 a month billed annually, and the free plan does not include it. Google's plan page lists Deep Research in Gemini among paid plans and gives no numeric limit. Perplexity says Deep Research has been free for all users since its February 2025 launch, with a daily cap for non-subscribers and higher volume for Pro. The current cap is not published on the pages we checked.

ChatGPT's free deep research allowance was set in 2025, and reports of the number vary. We did not confirm the current figure on OpenAI's pages, so check its help center before you rely on it. For a single report, choose the agent by its source base and its checks, then pay for the tier that covers your monthly volume.

Checklist

List the sources the report must draw on before you pick an agent.
Run the same brief through two agents and compare their citations.
Open five cited claims in each report and check them against the source.
Compare plan limits on each vendor's own pricing or help page.
Record the date and version of each test so you can repeat it.

Questions & answers

Which deep research agent scored highest on DRACO?

In the October 2026 run published by Hyperresearch, Hyperresearch Deep scored 77.7, the highest mean. Claude Research (Opus) scored 73.0, while ChatGPT Deep research scored 55.2 and Gemini Deep Research 54.7. Perplexity Deep Research is listed at 57.8 on 8 tasks. The run is one company's measurement, so repeat it on your own questions.

Are deep research agents safe for client work?

Only with checks. Any agent can misquote a figure or lean on a weak source. Verify the numbers that matter against the original, keep a record of the sources, and disclose AI use wherever your client or employer requires it.

Sources for this entry

Vendor descriptions are attributed to the vendor. Prices change; confirm with each vendor.

Continue this reading path

Cite this entry

Best AI for Research. “Deep Research Agents Compared.” Hyperresearch reference, reviewed 7 November 2026.
https://bestaiforresearch.com/deep-research-agents-compared/