Skip to content
arxiv

Quick start

Run your first arxiv command.

Once arxiv is on your PATH:

arxiv search "attention" --cat cs.CL

That prints a table if you are at a terminal and NDJSON if you are piping. No flag needed either way.

One paper

arxiv paper 1706.03762

One request, and you get the id in both styles, the version, both DOIs, the title, the abstract, the comment, the journal reference, the authors and the categories.

arxiv paper 1706.03762 --depth full

Four requests, and you also get the version history with sizes and source types, the submitter, the licence, the report number, the subject classes and the full category names. It takes about twenty seconds, because the fourth request is on arxiv.org and the tool waits fifteen seconds before touching that host. See depth for what each level costs and buys.

Look before you fetch

Five commands answer from tables the binary carries, with no request at all.

arxiv fields                  # every field on a paper, where it comes from, what it costs
arxiv fields --depth quick    # what one request actually gets you
arxiv surfaces                # the twelve places this tool reads
arxiv routes                  # every URL it will ever request
arxiv grammar                 # the query language, with examples that run
arxiv planes                  # the pace it keeps, and why

arxiv id is the sixth, and it classifies a reference before a request goes out.

arxiv id hep-th/9711200
arxiv id 10.48550/arXiv.1706.03762
arxiv id https://arxiv.org/abs/2401.00001v3

Browsing

arxiv categories                  # every category code and name
arxiv category cs.CL              # one category, with its group and archive
arxiv list cs.CL 2026-01          # a month of a category
arxiv new cs.CL                   # today's announcements, with the announce type

Piping

Every command prints NDJSON to a pipe, so jq works with no flag.

arxiv search "transformer" --cat cs.LG -n 50 | jq -r '.id + "  " + .title'
arxiv search "diffusion" --from 2026-01 --fields id,title,submitted -o csv > papers.csv

--fields takes a comma separated allowlist and works with every format.

Next

2 min read