Undermind

The homepage describes a four-step flow: you describe the research goal, Undermind asks follow-up questions, searches the literature while following citation trails, and then lets you iterate on reports and open full texts. The FAQ puts a typical search at about 2 minutes, because it works in stages rather than returning one ranked list, and says the report summary will tell you when the search has probably not found everything yet so you can extend it with more compute.

The source is the Semantic Scholar database of over 200 million articles. If a specific paper is not in that database it will not appear, whatever the search does.

Depth, mainly. The FAQ says the Free tier searches abstracts and metadata, and that Pro adds compute to search the full texts of nearly all open-access and preprint articles. The pricing section lists the same AI models for chat, deep searches and reports on Free, then adds “the latest, most powerful AI models”, deeper full-text analysis, 10x higher usage limits and unlimited workspaces and paper libraries to Pro. No page gives a number for Free’s “standard rate limits”, so how many deep searches a free account gets is unknown.

The pricing section has separate Industry and Academic tabs and a monthly or annual switch, and the Pro price changes with each. Industry Pro is $75 monthly or $60 a month billed annually; Academic Pro is $20 monthly or $16 a month billed annually. The pricing page and homepage do not say how Academic status is checked, so confirm eligibility before paying. Team is per person and Enterprise is custom, with the terms noting that organisations may be under a separate written agreement.

Yes. The MCP page lists tools that launch a deep search, search papers by author, journal or year, read full-text PDFs in parallel, and write cited Markdown notes into your workspace. Open-access PDFs are fetched automatically; paywalled ones must be uploaded first. Because the terms make you responsible for having the right to upload or retrieve content through your own subscriptions, that upload step is a licensing decision for you, not a setting.

The homepage benchmark compares Undermind with three agentic web-search setups on 23 research goals drawn from its own users’ queries, and reports 85% of the 20 key papers found at 10 minutes against 50%, 47% and 43%. The vendor ran and described it, and we did not reproduce it. Its own terms also say outputs may be inaccurate and must be verified before reliance, so treat the ranked list as a starting set to check, not a complete record. See the refund policy before buying: charges are non-refundable once a billing period begins.

1. Pick a research question in a field you know and list five relevant papers, at least two of them open-access and two older than five years.
2. Describe the question in plain language on a free account and let the deep search finish; note how long it takes and whether the report summary says the search has converged.
3. Check which of your five papers appear, and whether each cited statement links to the paper it claims to come from.
4. Count how many deep searches the free account allows before it stops, since no page publishes the limit.
5. Note whether the results mention full-text details, which the FAQ says Free does not search.

Accept if at least four of the five known papers appear and the cited statements trace back to the right sources. Reject if known papers are missing without the summary saying the search is incomplete. No result is recorded here.