Ai2 Asta vs Elicit: 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.

Elicit is made by Elicit Research, PBC, an AI company that grew from Ought and focuses on scientific research and evidence-backed decisions. Its product searches academic literature, organizes papers into tables, extracts information, and guides systematic reviews. Elicit announced a $22 million Series A in February 2025 at a reported $100 million valuation, led by Spark Capital and Footwork. The company said it had more than 400,000 monthly researchers at that time, while its homepage later claimed more than five million researchers. Those figures use different dates and definitions. Elicit is active and has expanded from paper search into reports, routines, libraries, and organizational research workflows. This makes it useful for recurring work where teams need a consistent structure for reviewing literature and recording study details.

A normal Elicit workflow begins with a natural-language research question. Search returns relevant papers, after which the user can inspect titles, abstracts, full text, and extracted fields in a table. The user can add columns describing population, intervention, outcome, method, or another question, then export results. Research Reports automate search, screening, extraction, and report writing around a selected question and template. Systematic Reviews provide setup, paper gathering, title and abstract screening, optional full-text screening, data extraction, and a final report. Users can upload PDFs, import from Zotero, save papers in a library, chat with papers, and use a browser extension for institutional full-text access. In practice, the table format helps reviewers compare studies without moving repeatedly between separate documents or tools.

Elicit is strongest for researchers who need structured literature work rather than a general chatbot answer. It can reduce repetitive screening and extraction, but generated fields still need checking against papers, especially for nuanced methods, contradictory findings, and incomplete full text. Compared with Consensus, Elicit offers deeper table extraction and systematic-review workflows. Compared with SciSpace, it emphasizes structured review automation and research reports. Compared with NotebookLM, it searches a large paper corpus rather than relying mainly on user-uploaded sources. Choose Elicit for repeatable literature reviews, academic teams, and research operations; choose a general assistant for broad web research or creative writing. Its main tradeoff is that users gain review structure while accepting plan limits and the need for expert judgment.