Consensus vs Undermind: pricing, features and which to choose

Consensus is made by Consensus NLP, Inc. and presents itself as an AI search engine for research. Its documentation says it searches more than 400 million scholarly articles, including 350 million peer-reviewed papers, while an earlier company primer cited more than 200 million papers. The company says more than five million researchers, students, and clinicians use it across more than 10,000 universities. It is best understood as a retrieval and explanation layer over scholarly literature, rather than a general chatbot. Its defining feature is traceability: answer sentences link to papers, and users can inspect supporting quotations. The company does not publish a founding year or funding total in the material described here, so its identity rests mainly on product scope, audience, and scholarly-search positioning.

A typical workflow begins with a natural-language research question. Consensus searches its paper database, retrieves relevant results, and generates a response from those results rather than relying only on general model knowledge. Users can inspect citations, run Pro searches, and open Study Snapshot entries covering details such as methodology, sample size, and study duration. Pro analysis summarizes up to 20 papers, while Deep Review searches across up to 50 papers for broader literature work. The service also flags retracted papers and supports API and MCP connections. This workflow helps users form reading lists, compare findings, screen evidence, or gain orientation in an unfamiliar field. It is most useful before close reading, because important claims still require checking the full papers, methods, limitations, and surrounding evidence.

The main limitation is dependence on indexed scholarly metadata and AI summaries. Citation links make checking easier, but they cannot guarantee that a synthesis captures study quality, conflicting findings, publication bias, or methodological nuance. Consensus is less suited to citation-network exploration than ResearchRabbit, Connected Papers, or Litmaps, and less suited to full reference management than Elicit or Scite. Its 2026 privacy policy says it does not sell personal information or use queries, uploads, or personal data to train large language models, while still describing account, cookie, provider, and analytics processing. Choose Consensus for fast, citation-linked answers and evidence comparison; choose ResearchRabbit for visual discovery and iterative paper mapping. It best fits students, clinicians, researchers, and teams seeking orientation before deeper evaluation.

Undermind is an AI research agent for scientific literature, founded in 2023 by quantum physics PhDs Joshua Ramette and Tom Hartke. The company joined Y Combinator’s Summer 2024 batch and is based in San Francisco. Its target users include research scientists, doctors, and R&D teams facing narrow questions that keyword search can miss. Undermind says scientists from thousands of research institutions use the service, and its enterprise page identifies more than 1,000 GSK scientists as users. The founders describe the product as a search engine rebuilt around adaptive exploration, with large language models directing searches through scientific literature instead of matching terms alone. The company does not publish a funding total.

A typical search begins with a natural language research goal. Undermind asks follow-up questions, then reads and evaluates papers, follows citation trails, and continues exploring related authors and work. Users can iterate on reports, inspect full texts, extract details, generate custom tables, and trace statements through inline citations to source papers. Workspaces support shared research, paper libraries, and saved files. The service also offers alerts for new findings in selected areas. The company says a production deep search stops after about 2.9 minutes on average. That delay is deliberate: Undermind is designed for difficult discovery and evidence gathering, not for instant answers to routine questions.

Undermind fits researchers who value broad discovery and citation trails over speed. Elicit is a stronger choice for structured extraction and paper comparison, Consensus is better for concise evidence summaries, and SciSpace is better suited to explaining individual PDFs. Undermind is also a poor fit for general web research or teams seeking a self-serve public API, because API access is presented as an enterprise feature. Individual terms restrict use of the site and functionality to personal, noncommercial use, although insights may be used in commercial or professional research. Enterprise terms address proprietary data, model training, provider retention, and deletion. Choose Undermind for complex scientific discovery, then verify important findings against the cited papers. Its pricing separates academic subscriptions from commercial research subscriptions.