Undermind Review: Deep Search Agent for Scientific Literature [WTF Radar]

Scorecard for Undermind on a dark navy background

Undermind is a YC-backed AI agent that searches thousands of scientific papers to find the relevant ones, reads full texts and monitors new work. How it works, its benchmark claims, pricing and alternatives.

Most research agents summarize the top sources they find. Undermind is built for the opposite job: making sure you did not miss the paper that matters.

Key Takeaways

  • Built for recall. — Deep search reviews thousands of papers and brings the relevant ones into a workspace, stopping after about 2.5 minutes on average.
  • Self-reported lead. — Undermind claims 73% recall against 29 to 39% for frontier-model agents on its own 23-goal benchmark.
  • Cheap and connectable. — Free plan, Pro at $16 a month billed annually, and MCP connections for ChatGPT and Claude.

What does Undermind do?

Undermind launched in August 2024 out of Y Combinator's Summer 2024 batch as an agent that mimics how a human researcher searches, adapting queries as it reads (Product Hunt launch). Its deep search reviews thousands of papers, fetches open-access PDFs, reads full texts and figures, reviews evidence quality and writes with citations. Searches stop on their own after about 2.5 minutes on average, and you can monitor a topic for new publications (Undermind).

Undermind says it is used by scientists at thousands of institutions, by more than 1,000 GSK scientists, and by nine U.S. Department of Energy national labs. Its headline claim is twice as many relevant papers as AI web search, and its own benchmark reports 73% recall and 60% precision at ten minutes against 29 to 39% recall for frontier-model agents. It connects to ChatGPT and Claude through MCP.

How does Undermind score?

Editorial scores from published documentation and announcements through October 31, 2026, not benchmark results. Total: 14/20, on the same criteria as the Best AI Research Tools guide.

CriterionScoreWhy
Depth4/5Adaptive search that changes queries based on what it finds; reads full texts and figures.
Citations4/5Paper-level sources with evidence quality review; benchmark is vendor-run.
Reach2/5Scientific literature; open-access PDFs plus your uploads.
Access4/5Free plan, low-cost Pro and Team, MCP access.

How is Undermind priced?

As published in October 2026: Free for exploratory research; Pro at $16 a month billed annually, with the newest models, deeper full-text analysis and ten times the usage limits; Team at $15 per person a month billed annually; Enterprise custom (Undermind pricing).

Who is Undermind for?

  • Fits: scientists and R&D teams starting a project or review who need every relevant paper, not a summary of a few.
  • Look elsewhere: for structured extraction and PRISMA reviews (Elicit), quick evidence checks (Consensus) or non-scientific questions (Gemini Deep Research).

What are the Undermind alternatives?

Sources

What is Undermind?

Undermind is an AI search agent for scientific literature, built by a Y Combinator Summer 2024 company. You describe a research question in plain language; it searches iteratively, reads papers and returns the ones that are relevant, with tools to monitor new publications.

Is Undermind better than Google Scholar?

Undermind claims it finds far more relevant papers than keyword search and reports 73% recall on its own benchmark. That benchmark is vendor-run and has not been independently checked. For complex questions that keyword search handles poorly, it is worth trying side by side.

How much does Undermind cost?

As published in October 2026, there is a free plan, Pro at $16 a month billed annually with ten times the free usage limits, Team at $15 per person a month billed annually, and custom Enterprise pricing.

Explore More

Ready to Find the Right AI Tools?

Browse our data-driven rankings to find the best AI tools for your team.