PaperQA AI

open_source

PaperQA2 is an open-source AI agent by FutureHouse that surpasses PhD-level researchers in scientific literature search, retrieval, and summarization tasks.

About

PaperQA2, developed by FutureHouse (a 501(c)(3) nonprofit), is the first AI agent to demonstrate superhuman performance across a range of scientific literature search and synthesis tasks. Evaluated using LitQA2 — part of the LAB-Bench benchmark — PaperQA2 outperforms PhD and postdoc-level biology researchers in retrieving accurate information from scientific papers. The agent is equipped with tools to search and retrieve papers, extract relevant content, traverse citation networks, and formulate well-grounded answers. Building on PaperQA2, FutureHouse created WikiCrow, which generates Wikipedia-style scientific summaries more accurate than human-curated Wikipedia articles as judged by expert reviewers. It has already been used to produce articles covering all 20,000 genes in the human genome, synthesizing over 1 million distinct papers. Another application, ContraCrow, identifies contradictions between published scientific papers, uncovering an average of 2.34 contradicted statements per paper — a powerful tool for hypothesis generation and research prioritization. PaperQA2 is open-source and available on GitHub, making it accessible to researchers, developers, and institutions looking to perform literature analysis at a scale that is currently infeasible for human researchers. It is ideal for biomedical research, systematic reviews, knowledge base construction, and automated scientific writing.

Key Features

Superhuman Literature Retrieval: Achieves higher accuracy than PhD and postdoc-level biology researchers on the LitQA2 benchmark for retrieving information from scientific papers.
Citation Graph Exploration: Traverses citation networks to find relevant connected papers, enabling deep, contextual synthesis of scientific knowledge.
Wikipedia-Style Summary Generation (WikiCrow): Generates accurate scientific summaries at scale — proven more accurate than human-curated Wikipedia articles when evaluated by expert biologists.
Contradiction Detection (ContraCrow): Evaluates every claim in a scientific paper to identify contradicting evidence elsewhere in the literature, averaging 2.34 contradictions per paper.
Massive-Scale Literature Analysis: Capable of synthesizing information from millions of papers simultaneously — demonstrated by generating gene summaries across all 20,000 human genome genes.

Use Cases

Biomedical researchers using PaperQA2 to rapidly retrieve accurate answers from thousands of scientific papers without manual literature review.
Academic institutions automating the generation of comprehensive, Wikipedia-style gene or disease summaries using WikiCrow for large-scale knowledge bases.
Research teams using ContraCrow to identify contradictions between studies, surfacing gaps and opportunities for novel hypothesis generation.
Developers and data scientists building custom scientific Q&A or summarization pipelines on top of the open-source PaperQA2 framework.
Systematic review authors leveraging PaperQA2 to accelerate evidence synthesis across hundreds of studies in a fraction of the time.

Pros

Superhuman Accuracy: Benchmarked to outperform PhD/postdoc-level researchers on scientific retrieval tasks, providing a highly reliable research assistant.
Open Source & Free: Fully open-source and backed by a nonprofit, making it freely accessible to researchers, developers, and academic institutions.
Scales to Millions of Papers: Can synthesize information from over a million papers simultaneously, enabling analyses impossible for human researchers.
Multi-Purpose Research Applications: Supports diverse use cases including Q&A, summary generation, contradiction detection, and hypothesis generation.

Cons

Primarily Focused on Biology: Current benchmarks and demonstrations are centered on biology; performance on other scientific domains may vary.
Requires Technical Setup: As an open-source tool available via GitHub, non-technical users may find deployment and configuration challenging without developer support.
Dependent on Available Literature: Output quality is constrained by the accessibility and coverage of the scientific papers it can retrieve and index.

Frequently Asked Questions

PaperQA2 is an open-source AI agent developed by FutureHouse that performs superhuman scientific literature search, retrieval, and summarization, outperforming PhD-level researchers on standardized benchmarks.

It uses a suite of tools to find papers, extract information, explore citation graphs, and formulate answers — all evaluated against LitQA2, a rigorous scientific retrieval benchmark where it outperforms expert human researchers.

Yes. PaperQA2 is fully open-source and available on GitHub. FutureHouse is a registered 501(c)(3) nonprofit, and the tool is freely available to the research community.

WikiCrow is an agent built on top of PaperQA2 that generates Wikipedia-style summaries of scientific topics. Its summaries have been shown to be more accurate on average than existing Wikipedia articles, as judged by blinded PhD-level reviewers.

ContraCrow is another PaperQA2-based agent that detects contradictions between published scientific papers. It evaluates claims in a paper against the broader literature, finding an average of 2.34 contradicted statements per paper, which can guide new hypothesis generation.