Free AI Literature Search for arXiv

Write one or two sentences the way you'd explain the topic to a colleague: what you're studying, the method or setting you care about, and anything you want left out. Any language works; the searches themselves run in English. The limit is 300 characters.

Keyword searches catch papers that use the exact terms, acronyms or model names you'd expect. Semantic searches match by meaning, so they also find papers that describe the same idea in different words. Each search covers a different angle, and the results show which searches found each paper.

The free search is a fast scan of a topic, not an exhaustive review. Each search runs at most five search routes and takes at most 200 papers from each, so up to 1,000 candidates are merged and screened. A broad topic can match far more papers than that, so some relevant ones may not appear; narrowing the question usually helps more than rewording it. There is also a small daily allowance of searches per network, and results stay available at their link for 30 days.

Without an account, the page lists the top 20 papers by inclusion probability, plus the top 20 for each topic or search route you select. A free account lets you page through every included paper in each view and save the results as a collection to keep working on them.

Yes. Write it into your question, such as "since 2023", "with released code" or "NeurIPS and ICML papers", and each search is narrowed before papers are picked. Venues cover computer-science conferences and journals (from DBLP) and only papers already formally published, so recent preprints drop out. Venues outside computer science, citation counts and impact factors aren't available; the search runs on the topic and says that limit wasn't applied.

After the search, AI writes inclusion and exclusion criteria from your question and the papers it found. Jev, our screening model, reads each title and abstract against them and gives an inclusion probability; papers at 50% or above are included. It's a screening aid, not a verdict: the criteria are shown with the results, so check key papers yourself.

Usually one of four reasons: it isn't on arXiv; it didn't rank in the top 200 of any of the (at most five) searches, which narrowing the question or adding its key terms often fixes; it was screened out by the criteria shown with the results; or it is included but outside the top 20 shown without an account.

arXiv only: physics, mathematics, computer science, quantitative biology and finance, statistics, electrical engineering and economics. Journal-only fields such as clinical medicine or law are outside it.