Ai2 Asta vs Consensus: pricing, features and which to choose

Ai2 Asta is a web-based scientific research assistant from the nonprofit Allen Institute for AI, launched publicly in 2025. Ai2 describes Asta as part of a broader initiative to build trustworthy agents for scientific work. The service combines scholarly search, cited report generation, dataset analysis, automated discovery, and theory generation. Its product page says the literature index includes more than 108 million abstracts and 12 million full-text papers. Asta targets researchers, students, analysts, instructors, and science communicators rather than general-purpose writing users. Ai2 does not publish funding figures, a founding date for the product, or a separate corporate entity for Asta. The hosted service runs on the web, and signed-in users can preserve questions and results.

Asta puts several research tasks into one browser workflow. Find papers searches its scholarly index, while Generate a report produces a cited synthesis that users can inspect and extend. Analyze data accepts structured files and returns explanations, visualizations, copyable code, assumptions, statistical tests, and a methods section. Follow-up prompts add analysis to the report while retaining prior work. AutoDiscovery runs experiments designed to surface patterns, and Theorizer retrieves source papers before generating candidate theories. Ai2 says theory runs usually take 20 to 30 minutes, while comprehensive runs can exceed an hour. Users must read the cited papers and check each claim because a generated theory is a research lead rather than a verified conclusion.

Asta's main limitation is the review burden attached to its most ambitious outputs. Ai2 says generated theories are not necessarily true, and its FAQ reports that about half of predictions had a later paper available for checking. Hosted users should avoid uploading sensitive or regulated datasets. Ai2 says raw datasets are deleted within seven days and are not used for model training, while prompts and interactions may be logged unless users opt out. Semantic Scholar is better for scholarly search and metadata APIs. Elicit is stronger for structured evidence review, and Consensus focuses on fast paper-based answers. Asta covers more exploratory analysis than those products. Researchers wanting free cited literature work with small-dataset exploration should pick Asta.

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.