Questions & answers

Hyperresearch is our editorial pick for accuracy in long reports. In its own October 2026 DRACO run it scored 77.7 on 10 tasks, with the same judge for every tool, against 73.0 for Claude Research (Opus). Benchmarks from vendors are a starting point, so check the sources in any report yourself. Elicit suits paper-based questions better, and Perplexity suits quick sourced lookups.

Not always. A free plan can cover a few trial questions, and the first Deep report from Hyperresearch is free. A paid plan makes sense when you run research every week, need longer reports, or need team features. Work out how many reports you expect in a month, then compare that with the limits on each plan before you pay.

Elicit's pricing page lists a free Basic plan. Its Pro plan is $49 per user, and the page shows $588 per user when billed annually. Pro screens up to 5,000 papers. The page does not state monthly billing separately, so confirm the terms before you sign up. Check the research limits on the plan you choose, since the page does not give those numbers.

Consensus's pricing page lists Individual, Team and Enterprise plans but shows no prices for them when we checked. Third-party sites give figures that disagree with each other, so we do not quote one. Check the pricing page or contact the vendor for a current quote. Ask about research limits and paper collection sizes at the same time, since both affect what you get for the price.

Google offers a free Gemini plan, but its pricing page does not publish a numeric Deep Research cap on any plan. Paid plans listed there include Google AI Plus at $4.99 a month, Google AI Pro at $19.99 a month and Google AI Ultra from $99.99 a month. Test your usual research task on the free plan first, then decide whether the paid plan's limits suit your work.

Elicit is built around literature review, and its Pro plan screens up to 5,000 papers on its own pricing page. Scite suits work where you need to see how papers cite one another, and its Pro plan lists 2,500 credits a month. Perplexity and ChatGPT are better for quick background reading than for systematic screening. Hyperresearch is our pick for sourced reports, not paper screening.

Yes, if you check the terms first. Read the data retention and training policy for the plan you plan to use, and confirm whether team features such as admin controls and SSO are included. Do not upload confidential client files to a personal account before you have checked the policy. Get any important answer in writing from the vendor, because a support chat reply may not count as a policy.

It depends on the vendor, the plan and the setting you choose. Some consumer plans let you turn training off, while many business plans exclude training by contract. Policies change, so read the current version for your exact plan. If the answer matters for your work, ask the vendor to confirm it in writing before you upload any sensitive material.

Pick a question you already know well, then send the same brief to each tool. Count the sources, open five at random and check the claims. Note every limit you hit and every figure with no source. Use a free option where one exists, and move to a paid trial only for the tool that performs best. The cost per report that passes your checks is the number to compare.

Scores depend on the task set, the judge, the number of tasks completed and the version tested. A tool that finished 8 of 10 tasks can score differently from one that finished all 10. Vendors also publish their own runs. Read the method, check how many tasks were completed, and compare tools only when they were scored on the same set.

Not always. Aggregator sites often disagree with each other and with the vendor's own page. For example, sites list different Perplexity Pro daily caps, and different SciSpace Premium prices. Where a vendor's page does not show a number, we mark the figure as an estimate or leave it out. Check the official pricing page before you buy, and confirm any figure you plan to budget around.

No. A research tool speeds up finding and reading sources, and it can draft a useful first report. A person still needs to check the claims that matter, decide what counts as evidence and take responsibility for the output. Treat the tool as a fast assistant with a checklist, and keep a human review step for any decision with financial, legal or medical consequences.