PaperQA2

FutureHouse designed PaperQA2 to answer scientific questions the way a careful researcher would: search a corpus of papers, gather and re-rank evidence with contextual summarization, then compose an answer with grounded in-text citations pointing at specific sources. The agentic loop decides which tools to invoke at each step, and the team's research reports superhuman performance against human experts on scientific question answering, summarization, and contradiction detection benchmarks. It installs as pip install paper-qa on Python 3.11 or newer, offering both a pqa command-line interface and a Python API, and the project switched to calendar versioning in December 2025 as development continues. Model access goes through LiteLLM, so OpenAI, Claude, and Gemini models work alongside fully local ones served by Ollama or llama.cpp, with optional Crossref, Unpaywall, and Semantic Scholar keys easing large-scale indexing. The code is Apache-2.0 licensed and the repository has more than 9,000 stars. Research groups and biotech teams embed it wherever literature review needs to be fast, cited, and reproducible.