Undermind is an AI research agent that searches the scientific literature in depth: it asks what you need, reads and ranks hundreds of papers over about two minutes, follows citation trails and writes a report with every claim linked to a source. It draws on Semantic Scholar's database of over 200 million articles. The free plan searches abstracts and metadata with standard rate limits. Pro adds full-text search of open-access papers and preprints, the newest models and 10x higher limits, for $16 a month billed yearly for academics ($20 monthly) or $60 for industry ($75 monthly). Team plans are priced per person, and Enterprise adds organization-wide login and custom security terms.
Reviewed by EdGrows·Updated Oct 5, 2026This analysis is based on documentation, public user reports, and vendor materials — not yet on our own hands-on testing. How we rate
A typical search engine returns results in under a second, ranked from best to worst. Undermind takes about two minutes, on purpose. It behaves like a careful researcher: it asks follow-up questions about what you need, reads hundreds of candidate papers, follows citation trails and keeps going until it judges that it has found the important work.
The company was started by two quantum physics PhDs from MIT and is backed by Y Combinator. Its users include academic labs and pharmaceutical R&D teams, and its homepage says more than 1,000 scientists at GSK use it.
You start by describing your research in plain language, the way you'd explain it to a colleague. Undermind may ask clarifying questions first. Then the search runs in stages. It looks at the most promising papers, learns from what it finds and adjusts its strategy before the next pass.
The distinctive part is that it keeps score of itself. Undermind models statistically how much of the relevant literature it has probably found, and the report says so when a search hasn't converged. You can then extend the search with more computation to find the rest. A missing paper usually means the search needs more time, or occasionally that the paper isn't in the database at all.
Results arrive as a report in which each statement links back to its source paper. From there you can chat about the findings, generate tables of details pulled from the papers, filter results by relevance and set alerts for new publications on the topic.
The academic price list, billed annually, on October 5, 2026. An Industry tab sits beside it.
| Free | Pro | Team | Enterprise | |
|---|---|---|---|---|
| Academic, billed yearly | $0 | $16 a month | $15 per person a month | Custom |
| Academic, billed monthly | $0 | $20 | $18.70 per person | Custom |
| Industry, billed yearly | $0 | $60 a month | $60 per person a month | Custom |
| Industry, billed monthly | $0 | $75 | $75 per person | Custom |
| What it adds | Chat, deep searches and reports with standard limits, shared workspaces, agent connections | Newest models, full-text analysis, 10x higher limits, unlimited workspaces, files and libraries | Member management, priority support, central billing | More compute, organization-wide login, onboarding, admin dashboard, custom terms |
Yearly billing saves 20%. The Team price is per person, and the page shows it for five seats.
Companies get stronger privacy commitments on Enterprise. Undermind says R&D data stays isolated and is never used to train models, its model providers don't retain it, and customers own their inputs and outputs.
The free plan is a real search engine, not a teaser. It runs deep searches and writes reports, but it reads only abstracts and metadata. Pro spends more compute to search the full texts of nearly all open-access articles and preprints. That matters when the detail you need sits in a methods section or a supplementary table rather than in the abstract.
Pro also raises usage limits tenfold and uses Undermind's newest models. For a PhD student running a literature review over several months, the academic Pro price of $16 a month billed yearly is modest. For a company, $60 a month per researcher is a budget line, and the free plan is the place to judge whether the extra papers it finds are worth it.
Everything Undermind finds comes from Semantic Scholar's index of more than 200 million articles, which includes PubMed and arXiv. That covers most of science well. It can still miss papers that aren't indexed there, and paywalled articles are searched by abstract rather than full text.
Treat the homepage benchmarks with care. Undermind reports that its first search engine beat Google Scholar by a factor of ten, and that its current version outperforms frontier AI agents using web search. These are the company's own tests. They're a reason to try it, not proof it fits your field.
Undermind runs an MCP server, so assistants such as Claude and ChatGPT can call its searches directly. Coding agents can use it the same way. A researcher can ask their usual assistant a question and have Undermind do the literature search behind it, with citations intact.
Undermind is strongest at the first, hardest step: finding the papers. Other tools take over after that. Elicit extracts data from a set of papers into tables for a systematic review. Scite tells you how a paper has been cited since, including by studies that disagree with it. ResearchRabbit maps how a set of papers connect through citations, and Consensus is faster when you just want to know what the evidence says. Our AI research tools guide explains how they divide the work.
Yes. The free plan includes chat, deep searches and reports with standard rate limits, shared workspaces and connections to assistants such as Claude and ChatGPT. It searches abstracts and metadata; reading full texts needs Pro.
Because it works in stages. A typical search takes about two minutes while it scans hundreds of papers, compares and ranks them and adjusts its strategy. It also estimates how much of the relevant literature it has found, and you can extend a search that hasn't finished.
Semantic Scholar's database of over 200 million articles, which includes PubMed, arXiv and many other sources across the sciences. Pro also reads the full texts of nearly all open-access articles and preprints.
It publishes separate price lists. Pro costs $16 a month billed yearly for academics and $60 for industry users, or $20 and $75 billed monthly. Enterprise contracts add organization-wide login, admin tools and custom security reviews.
Yes. Undermind runs an MCP server, so AI assistants and coding agents that support MCP, including Claude and ChatGPT, can run searches and use the results.
Cited AI search plus the new Computer agent: free plan, Pro at $20 or $200 a year, Max at $200, Computer credits, and the free Comet browser.
An academic search engine that answers from peer-reviewed papers. Basic search is free, Pro starts at $12 a month billed yearly, and students get up to 40% off.
A free chat app and one of the cheapest frontier-class APIs: V4.1-Flash with vision and open weights, from $0.15 per million input tokens off-peak.